Full Metrics by Condition

AI Condition Classifier Performance

Vetology’s AI classifiers have been validated using a foundation of over 300,000 multi-image patient cases from real-world veterinary practices. Below you’ll find transparent performance data for every classifier. Use the “details” dropdown to see even more in-depth data.

This deep, real-world dataset helps our AI support veterinarians with reliable screening results at the point of care.

Vetology is the ONLY veterinary AI imaging platform that offers this level of transparency.  We publish complete performance metrics to help you make informed decisions about diagnostic accuracy and clinical implementation.

While these results offer meaningful insight into expected performance, real-world factors such as image quality, positioning, and patient variability can influence accuracy in clinical settings.

AI Condition Classifier Performance Metrics

Every canine and feline condition classifier currently included in your monthly subscription. Search it the way you would describe a case.

Snapshot of 94 published condition classifiers as of . Metrics are recalculated when a classifier is released and/or retrained.

94
Classifiers published
Each met its sensitivity and specificity targets
9
New in September
Marked in amber below
31,309
Median cases checked
Half the classifiers were checked against more cases than this

Every classifier, sensitivity against specificity

657585955565758595Constipation (Canine): sensitivity 72.9%, specificity 90.8%Decreased serosal detail (Canine): sensitivity 70.4%, specificity 95.5%Diffusely distended colon (Canine): sensitivity 71.6%, specificity 81.0%Distended and malpositioned stomach (GDV) (Canine): sensitivity 90.5%, specificity 92.7%Distended stomach (Canine): sensitivity 70.7%, specificity 87.1%Enlarged spleen (Canine): sensitivity 78.6%, specificity 91.5%Gastric foreign material (Canine): sensitivity 72.7%, specificity 76.2%Gastric malposition (Canine): sensitivity 74.5%, specificity 85.1%Hepatomegaly (Canine): sensitivity 73.6%, specificity 97.3%Kidney stones (left kidney) (Canine): sensitivity 72.7%, specificity 97.2%Kidney stones (right kidney) (Canine): sensitivity 77.9%, specificity 95.2%Left kidney size change (Canine): sensitivity 65.2%, specificity 78.5%Liver mass (Canine): sensitivity 74.0%, specificity 94.1%Mineral in the gallbladder or biliary tract (Canine): sensitivity 76.1%, specificity 94.8%Mineral or metal opaque small intestinal material (Canine): sensitivity 69.7%, specificity 90.7%Pregnancy (mineralized fetal skeletons) (Canine): sensitivity 99.1%, specificity 97.8%Right kidney size change (Canine): sensitivity 69.0%, specificity 85.9%Segmental small intestinal distension (Canine): sensitivity 71.2%, specificity 90.2%Small liver (Canine): sensitivity 68.6%, specificity 90.2%Splenic mass (Canine): sensitivity 70.5%, specificity 93.3%Urocystoliths / urethroliths (Canine): sensitivity 73.7%, specificity 95.1%Diffuse small intestinal distension (Feline): sensitivity 61.1%, specificity 78.8%Diffusely distended colon (Feline): sensitivity 75.8%, specificity 90.4%Distended stomach (Feline): sensitivity 74.2%, specificity 89.9%Enlarged spleen (Feline): sensitivity 72.2%, specificity 90.9%Hepatomegaly (Feline): sensitivity 72.3%, specificity 93.0%Kidney stones (left kidney) (Feline): sensitivity 81.7%, specificity 98.1%Kidney stones (right kidney) (Feline): sensitivity 71.4%, specificity 94.6%Left kidney margin change (Feline): sensitivity 73.0%, specificity 87.9%Left kidney size change (Feline): sensitivity 71.7%, specificity 83.8%Liver mass (Feline): sensitivity 75.5%, specificity 92.6%Mid-abdominal mass (Feline): sensitivity 81.7%, specificity 73.4%Mineral or metal opaque gastric material (Feline): sensitivity 71.9%, specificity 91.3%Pregnancy (mineralized fetal skeletons) (Feline): sensitivity 100.0%, specificity 99.4%Right kidney margin change (Feline): sensitivity 69.1%, specificity 82.0%Right kidney size change (Feline): sensitivity 76.2%, specificity 84.5%Segmental small intestinal distension (Feline): sensitivity 71.3%, specificity 90.0%Small intestinal foreign material (Feline): sensitivity 72.6%, specificity 84.7%Urocystoliths / urethroliths (Feline): sensitivity 71.2%, specificity 92.1%Congenital lumbar vertebral anomaly (Canine & Feline): sensitivity 74.8%, specificity 89.0%Congenital thoracic vertebral anomaly (Canine & Feline): sensitivity 85.0%, specificity 98.7%Degenerative joint disease, coxofemoral joints and pelvis (Canine & Feline): sensitivity 72.9%, specificity 95.9%Hip dysplasia (Canine & Feline): sensitivity 80.9%, specificity 90.9%Intervertebral disc disease, lumbar spine (Canine & Feline): sensitivity 76.6%, specificity 87.6%Intervertebral disc disease, thoracic spine (Canine & Feline): sensitivity 72.6%, specificity 89.6%Periarticular proliferation, pelvis (Canine & Feline): sensitivity 85.0%, specificity 95.3%Spondylosis deformans, lumbar spine (Canine & Feline): sensitivity 78.5%, specificity 93.5%Spondylosis deformans, lumbosacral junction (Canine & Feline): sensitivity 78.7%, specificity 94.3%Spondylosis deformans, thoracic spine (Canine & Feline): sensitivity 79.4%, specificity 91.0%Spondylosis deformans, thoracolumbar junction (Canine & Feline): sensitivity 90.0%, specificity 94.5%Caudodorsal alveolar pattern (Canine): sensitivity 89.6%, specificity 92.1%Caudodorsal pulmonary nodules (Canine): sensitivity 73.1%, specificity 98.3%Cranial mediastinal mass (Canine): sensitivity 71.3%, specificity 89.9%Cranioventral alveolar pattern (Canine): sensitivity 75.9%, specificity 94.0%Cranioventral pulmonary nodules (Canine): sensitivity 73.4%, specificity 96.4%Dilated bronchi (Canine): sensitivity 70.8%, specificity 85.2%Enlarged heart (Canine): sensitivity 89.9%, specificity 92.6%Enlarged left atrium (Canine): sensitivity 75.3%, specificity 99.3%Enlarged left auricular appendage (Canine): sensitivity 85.0%, specificity 97.7%Enlarged main pulmonary artery (Canine): sensitivity 72.1%, specificity 90.2%Enlarged pulmonary arteries (Canine): sensitivity 73.7%, specificity 81.3%Enlarged pulmonary veins (Canine): sensitivity 82.8%, specificity 90.1%Heart base mass (Canine): sensitivity 71.2%, specificity 89.9%Heart failure pattern (Canine): sensitivity 95.9%, specificity 92.8%Hiatal hernia (Canine): sensitivity 75.6%, specificity 90.3%Left mediastinal shift (Canine): sensitivity 74.2%, specificity 74.4%Left-sided cardiomegaly (Canine): sensitivity 96.3%, specificity 98.2%Microcardia (Canine): sensitivity 81.1%, specificity 85.3%Miliary pulmonary pattern (Canine): sensitivity 79.7%, specificity 97.3%Narrowed intrathoracic trachea (Canine): sensitivity 69.6%, specificity 90.5%Narrowed or compressed bronchi (Canine): sensitivity 89.9%, specificity 90.9%Obscuring pleural effusion (Canine): sensitivity 76.6%, specificity 99.2%Perihilar infiltrate (Canine): sensitivity 91.3%, specificity 90.9%Pleural effusion (Canine): sensitivity 90.9%, specificity 99.6%Pulmonary edema (Canine): sensitivity 91.0%, specificity 95.1%Pulmonary masses (Canine): sensitivity 73.7%, specificity 98.4%Thoracic esophageal dilation (Canine): sensitivity 83.5%, specificity 96.0%Thoracic lymphadenopathy (Canine): sensitivity 71.4%, specificity 97.0%Underinflated lungs (Canine): sensitivity 71.9%, specificity 89.4%Bronchial pattern (Feline): sensitivity 73.2%, specificity 93.2%Caudodorsal alveolar pattern (Feline): sensitivity 72.9%, specificity 91.0%Cranial mediastinal mass (Feline): sensitivity 71.2%, specificity 91.2%Cranioventral alveolar pattern (Feline): sensitivity 71.7%, specificity 92.0%Diffuse interstitial pattern (Feline): sensitivity 72.1%, specificity 87.0%Enlarged heart (Feline): sensitivity 77.0%, specificity 92.4%Enlarged pulmonary vessels (Feline): sensitivity 71.3%, specificity 90.6%Enlarged thoracic lymph nodes (Feline): sensitivity 74.4%, specificity 92.5%Left-sided cardiomegaly (Feline): sensitivity 70.2%, specificity 92.9%Left-sided heart failure pattern (Feline): sensitivity 92.9%, specificity 94.7%Obscuring pleural effusion (Feline): sensitivity 86.2%, specificity 96.2%Pleural effusion (Feline): sensitivity 88.1%, specificity 99.3%Pulmonary edema (Feline): sensitivity 77.8%, specificity 96.4%Pulmonary masses (Feline): sensitivity 84.5%, specificity 94.7%Thoracic esophageal dilation (Feline): sensitivity 74.1%, specificity 92.7%Specificity %Sensitivity %
Canine Feline Canine and feline

Each mark is one classifier. Up and to the right is stronger on both measures. Hover or tap a mark for its name.

Species covered

CanineCanine: 6161FelineFeline: 4444BothBoth: 1111

94 classifiers in total. 11 of them cover both species and are included in the canine and feline counts.

Body system

ThoraxThorax: 4444AbdomenAbdomen: 3939Spine and MSKSpine and MSK: 1111

Click to filter results

Species
Body system
What changed
94 classifiers
Understanding these metrics What each number means, and what it does not

Sensitivity

How often a classifier reports a condition that is present. Think of it as the catch rate. A sensitivity of 89% means that in 89 of every 100 patients that had the condition, it was on the report. The remainder is what the classifier missed, so higher is better.

Specificity

How often a classifier leaves a condition off the report when the condition is absent. A specificity of 92% means that in 92 of every 100 patients that did not have the condition, it was not on the report. Higher specificity means fewer false alarms and less follow-up on findings that were never there.

Positive predictive value (PPV)

Of the cases where a classifier reported the finding, how many turned out to have it. A PPV of 80% means 8 in every 10 positive results were confirmed. PPV moves with how common a condition is in the cases checked at least as much as it moves with the classifier itself, which is why prevalence is published next to it on every row. On an uncommon condition a strong classifier can still show a low PPV.

Prevalence

The share of the cases checked that were positive for the condition. It is not a measure of how often you will meet the condition in your own patients. Our classifiers are built and scored on the images we hold, so a low number here can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it. Prevalence is also the main thing that moves predictive value, so the two are meant to be read together.

Area under curve (AUC)

One number for how cleanly a classifier separates positive cases from negative ones. 1.0 would be perfect and 0.5 would be no better than a coin toss. Above 0.85 is strong. Data scientists use it to compare classifiers against each other; it is not a number to apply to a single patient.

Positive likelihood ratio

How far a reported finding moves the odds that the patient has it. A ratio of 10 means the finding is 10 times more likely to be reported in a patient that has the condition than in one that does not. Unlike predictive value it does not move with how common the condition is, so it describes the classifier rather than the set it was checked on. Anything under 5 is a small shift. You will find it in each classifier's details.

Negative predictive value (NPV)

How often a classifier is right when it does not report a condition. This number runs high on any uncommon condition, whether or not the classifier is good at finding it, so a high NPV on its own is not evidence that the condition is absent. Sensitivity is what answers that, and each classifier's details show how many patients it missed.

Overall accuracy

The share of all cases, positive and negative, that a classifier got right, shown in each classifier's details. It sounds like the number that matters most, and on an uncommon condition it is the most misleading one here: a condition present in 1 of every 100 cases would score 99% accuracy by reporting nothing at all. Sensitivity and specificity are what separate the two halves.

Cases checked

The number of radiologist-labeled cases a classifier was scored against. Each classifier has its own set rather than a shared pool, running from 6,452 to 56,820 cases. The details for each row split that number into positive cases, where the condition was present, and negative cases.

True and false positives and negatives

The four ways a result can land, known to data scientists as the confusion matrix and shown in each classifier's details. True positives were reported and confirmed. False negatives were present and missed. True negatives were absent and correctly left off the report. False positives were reported and not confirmed.

Release/retrain date

When the version shown here went live. For some classifiers that is a first release and for others a retrain. Every figure in that row was measured on the version this date refers to, and the figures are recalculated at each retrain.

Every figure on this page was validated by Vetology's data science team, and we welcome external validation of that work. If you are a researcher who would like to examine these numbers, check them against your own cases, or propose a study with us, get in touch.

FAQs Questions we get about the classifiers and the screening report

How accurate are Vetology's AI classifiers?

It depends on the condition, which is why all 94 are listed above with their own figures rather than behind one headline number. Across the set the median sensitivity is 74.1% and the median specificity is 92.1%. Sensitivity runs from 61.1% to 100.0% and specificity from 73.4% to 99.6%. Find the condition you care about and read its row.

What conditions does Vetology screen for?

All 94 are in the table above, and you can search it or filter it by species and body system. By body system that is 44 thorax, 39 abdomen, 11 spine and musculoskeletal. By species it is 61 canine and 44 feline, 11 of which are validated for both. They run from heart and lung findings such as cardiomegaly, pleural effusion and alveolar patterns, through abdominal organ size and masses such as hepatomegaly and liver mass, to kidney stones and spondylosis deformans.

What is the screening report for?

It supports your own interpretation of the images. It is not a diagnosis, and a decision to start or withhold treatment should rest on everything the classifiers cannot see: the patient in the exam room, the history and presenting signs, the labwork, and your own expertise.

What does a classifier have to clear before it goes live?

Each classifier is scored against a golden set labeled by board-certified veterinary radiologists, and has to meet an internal sensitivity and specificity target. A classifier that misses its target goes back for more work rather than into the product, so this page is the set that passed.

What is not on this page?

These 94 classifiers are what we screen for. A condition that is not on this list was not assessed at all, and its absence from a screening report is not a negative result for it.

What should I make of a sensitivity in the 70s?

Sensitivity and specificity answer different questions. Specificity in the 90s means the classifier correctly identifies most of the cases where the condition is absent. Sensitivity in the 70s means a condition the classifier did not find narrows the question rather than settling it. That is why the screening report belongs alongside your own judgment rather than ahead of it.

Some of these PPVs look low. What does that mean for my cases?

Usually that the condition was uncommon in the cases we scored, rather than that the classifier is weak. It also means the figure here is not the figure for your case. If you already suspect a finding from the history, the signalment or the physical exam, the odds are higher before the classifier looks at anything, and a positive result for that finding is right more often than this page shows. Each classifier's details carry the positive likelihood ratio, which is the part that does not move with how common the condition is.

Do these figures depend on which views were taken?

Yes. A radiograph collapses a three-dimensional patient onto a flat image, so everything the beam passes through is summed together. A lateral carries no right versus left, and a ventrodorsal carries no dorsal versus ventral. Overlapping structures can add up to look like a lesion, and a real lesion can be lost against something beside it of the same opacity. That is why the standard is two orthogonal views, and three for the thorax: a nodule hidden in the dependent lung on one lateral is often visible on the opposite lateral once that lung re-expands. The figures here were measured on studies that carried the projections each finding needs, so on a study missing one, the classifier is working with less than it was measured with. That limit is not particular to AI. Any single projection carries the same ambiguity whoever is reading it, because the information is not in the image. Ask us if you have questions.

Do these figures apply to young patients?

They were measured in skeletally mature patients, and several of these findings have a normal counterpart in a growing one. Until about six months the liver normally extends past the costal arch, so a healthy puppy liver can look enlarged. Reduced serosal detail is the normal state in puppies and kittens rather than a sign of fluid. The immature heart is normally larger relative to the thorax, and vertebral heart score is unreliable until about six months. The thymus silhouettes the cranial cardiac margin on laterals until about four to six months and is easily taken for pleural fluid. In the skeleton, open physes appear as linear lucencies across the bone and unfused ossification centers as separate bony islands, both of which can read as fractures at the joints and metaphyses. Comparison is what settles most of these, laterality first and above all an age-matched littermate when one is available.

Measured performance of every Vetology AI radiograph classifier in production as of September 9, 2026. 94 classifiers, canine and feline. Columns: condition, sensitivity, specificity, positive predictive value, prevalence in the cases checked, area under curve, number of radiologist-labeled cases checked, and the date that version was released or retrained.
Condition ↕ Sensitivity ↕ Specificity ↕ PPV ↕ Prevalence ↕ AUC ↕ Cases checked ↕ Release/retrain date ↕ Details
FelineThorax
Bronchial pattern
73.2%
93.2%
 67.3% 16.05% 0.927 18,558

Full metric set

Overall accuracy
90.0%
Positive predictive value (PPV)
67.3%
Prevalence in the cases checked
16.05%
Area under curve (AUC)
0.927
Negative predictive value (NPV)
94.8%
Positive likelihood ratio
10.8
Positive cases checked
2,979
Negative cases checked
15,579
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive2,979 cases, 16.05% 2,181true positives 798false negatives
Negative15,579 cases 1,061false positives 14,518true negatives

18,558 cases were checked, prevalence was 16%.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. A positive result is about 11 times more likely in a patient that has this condition than in one that does not. About 7 in every 10 positive results were confirmed.

Not noted. This classifier found about 73 in every 100 patients that had this condition, and missed 798 of 2,979. About 95 in every 100 negative results were correct. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Caudodorsal alveolar pattern
89.6%
92.1%
 10.5% 1.02% 0.955 31,115

Full metric set

Overall accuracy
92.1%
Positive predictive value (PPV)
10.5%
Prevalence in the cases checked
1.02%
Area under curve (AUC)
0.955
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
11.3
Positive cases checked
318
Negative cases checked
30,797
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive318 cases, 1.02% 285true positives 33false negatives
Negative30,797 cases 2,440false positives 28,357true negatives

31,115 cases were checked, prevalence was 1.0%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 11 times more likely in a patient that has this condition than in one that does not. About 1 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 90 in every 100 patients that had this condition, and missed 33 of 318. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineThorax
Caudodorsal alveolar pattern
72.9%
91.0%
 3.1% 0.39% 0.911 37,040

Full metric set

Overall accuracy
90.9%
Positive predictive value (PPV)
3.1%
Prevalence in the cases checked
0.39%
Area under curve (AUC)
0.911
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
8.1
Positive cases checked
144
Negative cases checked
36,896
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive144 cases, 0.39% 105true positives 39false negatives
Negative36,896 cases 3,315false positives 33,581true negatives

37,040 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 8 times more likely in a patient that has this condition than in one that does not. About 3 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 105 of the 144 patients that had this condition, and missed 39. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 65 to about 80 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Caudodorsal pulmonary nodules
73.1%
98.3%
 21.8% 0.64% 0.957 16,349

Full metric set

Overall accuracy
98.2%
Positive predictive value (PPV)
21.8%
Prevalence in the cases checked
0.64%
Area under curve (AUC)
0.957
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
43.5
Positive cases checked
104
Negative cases checked
16,245
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive104 cases, 0.64% 76true positives 28false negatives
Negative16,245 cases 273false positives 15,972true negatives

16,349 cases were checked, prevalence was 0.6%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 43 times more likely in a patient that has this condition than in one that does not. About 2 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 76 of the 104 patients that had this condition, and missed 28. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 64 to about 81 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

Canine & FelineSpine and MSK
Congenital lumbar vertebral anomaly
74.8%
89.0%
 2.4% 0.37% 0.886 36,851

Full metric set

Overall accuracy
89.0%
Positive predictive value (PPV)
2.4%
Prevalence in the cases checked
0.37%
Area under curve (AUC)
0.886
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
6.8
Positive cases checked
135
Negative cases checked
36,716
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive135 cases, 0.37% 101true positives 34false negatives
Negative36,716 cases 4,025false positives 32,691true negatives

36,851 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other spine and MSK classifiers

This classifier is the highlighted mark. Gray marks are the other 10 spine and MSK classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 7 times more likely in a patient that has this condition than in one that does not. About 2 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 101 of the 135 patients that had this condition, and missed 34. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 67 to about 81 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

Canine & FelineSpine and MSK
Congenital thoracic vertebral anomaly
85.0%
98.7%
 61.3% 2.38% 0.978 22,365

Full metric set

Overall accuracy
98.4%
Positive predictive value (PPV)
61.3%
Prevalence in the cases checked
2.38%
Area under curve (AUC)
0.978
Negative predictive value (NPV)
99.6%
Positive likelihood ratio
64.9
Positive cases checked
533
Negative cases checked
21,832
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive533 cases, 2.38% 453true positives 80false negatives
Negative21,832 cases 286false positives 21,546true negatives

22,365 cases were checked, prevalence was 2.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other spine and MSK classifiers

This classifier is the highlighted mark. Gray marks are the other 10 spine and MSK classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 65 times more likely in a patient that has this condition than in one that does not. About 6 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 85 in every 100 patients that had this condition, and missed 80 of 533. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Constipation
72.9%
90.8%
 7.0% 0.94% 0.883 7,458

Full metric set

Overall accuracy
90.6%
Positive predictive value (PPV)
7.0%
Prevalence in the cases checked
0.94%
Area under curve (AUC)
0.883
Negative predictive value (NPV)
99.7%
Positive likelihood ratio
7.9
Positive cases checked
70
Negative cases checked
7,388
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive70 cases, 0.94% 51true positives 19false negatives
Negative7,388 cases 681false positives 6,707true negatives

7,458 cases were checked, prevalence was 0.9%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 8 times more likely in a patient that has this condition than in one that does not. About 7 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 51 of the 70 patients that had this condition, and missed 19. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 61 to about 82 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Cranial mediastinal mass
71.3%
89.9%
 2.7% 0.39% 0.894 36,272

Full metric set

Overall accuracy
89.9%
Positive predictive value (PPV)
2.7%
Prevalence in the cases checked
0.39%
Area under curve (AUC)
0.894
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
7.1
Positive cases checked
143
Negative cases checked
36,129
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive143 cases, 0.39% 102true positives 41false negatives
Negative36,129 cases 3,639false positives 32,490true negatives

36,272 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 7 times more likely in a patient that has this condition than in one that does not. About 3 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 102 of the 143 patients that had this condition, and missed 41. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 63 to about 78 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineThorax
Cranial mediastinal mass
71.2%
91.2%
 3.2% 0.41% 0.897 25,439

Full metric set

Overall accuracy
91.1%
Positive predictive value (PPV)
3.2%
Prevalence in the cases checked
0.41%
Area under curve (AUC)
0.897
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
8.1
Positive cases checked
104
Negative cases checked
25,335
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive104 cases, 0.41% 74true positives 30false negatives
Negative25,335 cases 2,228false positives 23,107true negatives

25,439 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 8 times more likely in a patient that has this condition than in one that does not. About 3 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 74 of the 104 patients that had this condition, and missed 30. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 62 to about 79 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Cranioventral alveolar pattern
75.9%
94.0%
 39.1% 4.81% 0.932 7,778

Full metric set

Overall accuracy
93.1%
Positive predictive value (PPV)
39.1%
Prevalence in the cases checked
4.81%
Area under curve (AUC)
0.932
Negative predictive value (NPV)
98.7%
Positive likelihood ratio
12.7
Positive cases checked
374
Negative cases checked
7,404
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive374 cases, 4.81% 284true positives 90false negatives
Negative7,404 cases 443false positives 6,961true negatives

7,778 cases were checked, prevalence was 4.8%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 13 times more likely in a patient that has this condition than in one that does not. About 4 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 76 in every 100 patients that had this condition, and missed 90 of 374. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineThorax
Cranioventral alveolar pattern
71.7%
92.0%
 3.6% 0.42% 0.909 34,419

Full metric set

Overall accuracy
91.9%
Positive predictive value (PPV)
3.6%
Prevalence in the cases checked
0.42%
Area under curve (AUC)
0.909
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
8.9
Positive cases checked
145
Negative cases checked
34,274
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive145 cases, 0.42% 104true positives 41false negatives
Negative34,274 cases 2,751false positives 31,523true negatives

34,419 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 9 times more likely in a patient that has this condition than in one that does not. About 4 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 104 of the 145 patients that had this condition, and missed 41. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 64 to about 78 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Cranioventral pulmonary nodules
73.4%
96.4%
 42.5% 3.49% 0.937 25,608

Full metric set

Overall accuracy
95.6%
Positive predictive value (PPV)
42.5%
Prevalence in the cases checked
3.49%
Area under curve (AUC)
0.937
Negative predictive value (NPV)
99.0%
Positive likelihood ratio
20.4
Positive cases checked
894
Negative cases checked
24,714
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive894 cases, 3.49% 656true positives 238false negatives
Negative24,714 cases 888false positives 23,826true negatives

25,608 cases were checked, prevalence was 3.5%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 20 times more likely in a patient that has this condition than in one that does not. About 4 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 73 in every 100 patients that had this condition, and missed 238 of 894. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Decreased serosal detail
70.4%
95.5%
 29.4% 2.61% 0.896 22,450

Full metric set

Overall accuracy
94.8%
Positive predictive value (PPV)
29.4%
Prevalence in the cases checked
2.61%
Area under curve (AUC)
0.896
Negative predictive value (NPV)
99.2%
Positive likelihood ratio
15.6
Positive cases checked
585
Negative cases checked
21,865
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive585 cases, 2.61% 412true positives 173false negatives
Negative21,865 cases 987false positives 20,878true negatives

22,450 cases were checked, prevalence was 2.6%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 16 times more likely in a patient that has this condition than in one that does not. About 3 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 70 in every 100 patients that had this condition, and missed 173 of 585. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

Canine & FelineSpine and MSK
Degenerative joint disease, coxofemoral joints and pelvis
72.9%
95.9%
 10.3% 0.64% 0.902 13,298

Full metric set

Overall accuracy
95.8%
Positive predictive value (PPV)
10.3%
Prevalence in the cases checked
0.64%
Area under curve (AUC)
0.902
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
17.9
Positive cases checked
85
Negative cases checked
13,213
Release/retrain date
Aug 2025

What the board-certified radiologist labels showed

Positive85 cases, 0.64% 62true positives 23false negatives
Negative13,213 cases 539false positives 12,674true negatives

13,298 cases were checked, prevalence was 0.6%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other spine and MSK classifiers

This classifier is the highlighted mark. Gray marks are the other 10 spine and MSK classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 18 times more likely in a patient that has this condition than in one that does not. About 1 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 62 of the 85 patients that had this condition, and missed 23. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 63 to about 81 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineThorax
Diffuse interstitial pattern
72.1%
87.0%
 14.2% 2.91% 0.894 21,025

Full metric set

Overall accuracy
86.5%
Positive predictive value (PPV)
14.2%
Prevalence in the cases checked
2.91%
Area under curve (AUC)
0.894
Negative predictive value (NPV)
99.0%
Positive likelihood ratio
5.5
Positive cases checked
612
Negative cases checked
20,413
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive612 cases, 2.91% 441true positives 171false negatives
Negative20,413 cases 2,663false positives 17,750true negatives

21,025 cases were checked, prevalence was 2.9%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 6 times more likely in a patient that has this condition than in one that does not. About 1 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 72 in every 100 patients that had this condition, and missed 171 of 612. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Diffuse small intestinal distension
61.1%
78.8%
 8.7% 3.22% 0.757 40,486

Full metric set

Overall accuracy
78.2%
Positive predictive value (PPV)
8.7%
Prevalence in the cases checked
3.22%
Area under curve (AUC)
0.757
Negative predictive value (NPV)
98.4%
Positive likelihood ratio
2.9
Positive cases checked
1,303
Negative cases checked
39,183
Release/retrain date
Feb 2026

What the board-certified radiologist labels showed

Positive1,303 cases, 3.22% 796true positives 507false negatives
Negative39,183 cases 8,324false positives 30,859true negatives

40,486 cases were checked, prevalence was 3.2%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 2.9 times more likely in a patient that has this condition than in one that does not. About 9 in every 100 positive results were confirmed, a consequence of that rarity. Anything below 5 is considered a small change, so a positive result supports the finding without confirming it.

Not noted. This classifier found about 61 in every 100 patients that had this condition, and missed 507 of 1,303. About 98 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Diffusely distended colon
71.6%
81.0%
 3.7% 1.00% 0.831 28,267

Full metric set

Overall accuracy
80.9%
Positive predictive value (PPV)
3.7%
Prevalence in the cases checked
1.00%
Area under curve (AUC)
0.831
Negative predictive value (NPV)
99.6%
Positive likelihood ratio
3.8
Positive cases checked
282
Negative cases checked
27,985
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive282 cases, 1.00% 202true positives 80false negatives
Negative27,985 cases 5,328false positives 22,657true negatives

28,267 cases were checked, prevalence was 1.0%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 4 times more likely in a patient that has this condition than in one that does not. About 4 in every 100 positive results were confirmed, a consequence of that rarity. Anything below 5 is considered a small change, so a positive result supports the finding without confirming it.

Not noted. This classifier found about 72 in every 100 patients that had this condition, and missed 80 of 282. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Diffusely distended colon
75.8%
90.4%
 14.5% 2.10% 0.898 9,433

Full metric set

Overall accuracy
90.1%
Positive predictive value (PPV)
14.5%
Prevalence in the cases checked
2.10%
Area under curve (AUC)
0.898
Negative predictive value (NPV)
99.4%
Positive likelihood ratio
7.9
Positive cases checked
198
Negative cases checked
9,235
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive198 cases, 2.10% 150true positives 48false negatives
Negative9,235 cases 885false positives 8,350true negatives

9,433 cases were checked, prevalence was 2.1%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 8 times more likely in a patient that has this condition than in one that does not. About 1 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 76 in every 100 patients that had this condition, and missed 48 of 198. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Dilated bronchi
70.8%
85.2%
 1.5% 0.31% 0.843 35,966

Full metric set

Overall accuracy
85.1%
Positive predictive value (PPV)
1.5%
Prevalence in the cases checked
0.31%
Area under curve (AUC)
0.843
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
4.8
Positive cases checked
113
Negative cases checked
35,853
Release/retrain date
Sep 2026

What the board-certified radiologist labels showed

Positive113 cases, 0.31% 80true positives 33false negatives
Negative35,853 cases 5,324false positives 30,529true negatives

35,966 cases were checked, prevalence was 0.3%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 5 times more likely in a patient that has this condition than in one that does not. About 1 in every 100 positive results were confirmed, a consequence of that rarity. Anything below 5 is considered a small change, so a positive result supports the finding without confirming it.

Not noted. This classifier found 80 of the 113 patients that had this condition, and missed 33. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 62 to about 78 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Distended and malpositioned stomach (GDV)
90.5%
92.7%
 0.9% 0.08% 0.951 27,939

Full metric set

Overall accuracy
92.7%
Positive predictive value (PPV)
0.9%
Prevalence in the cases checked
0.08%
Area under curve (AUC)
0.951
Negative predictive value (NPV)
100.0%
Positive likelihood ratio
12.4
Positive cases checked
21
Negative cases checked
27,918
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive21 cases, 0.08% 19true positives 2false negatives
Negative27,918 cases 2,032false positives 25,886true negatives

27,939 cases were checked, prevalence was 0.1%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 12 times more likely in a patient that has this condition than in one that does not. About 9 in every 1,000 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 19 of the 21 patients that had this condition, and missed 2. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 71 to about 97 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Distended stomach
70.7%
87.1%
 18.6% 4.00% 0.856 6,895

Full metric set

Overall accuracy
86.4%
Positive predictive value (PPV)
18.6%
Prevalence in the cases checked
4.00%
Area under curve (AUC)
0.856
Negative predictive value (NPV)
98.6%
Positive likelihood ratio
5.5
Positive cases checked
276
Negative cases checked
6,619
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive276 cases, 4.00% 195true positives 81false negatives
Negative6,619 cases 856false positives 5,763true negatives

6,895 cases were checked, prevalence was 4.0%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 5 times more likely in a patient that has this condition than in one that does not. About 2 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 71 in every 100 patients that had this condition, and missed 81 of 276. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Distended stomach
74.2%
89.9%
 10.6% 1.59% 0.915 16,403

Full metric set

Overall accuracy
89.7%
Positive predictive value (PPV)
10.6%
Prevalence in the cases checked
1.59%
Area under curve (AUC)
0.915
Negative predictive value (NPV)
99.5%
Positive likelihood ratio
7.4
Positive cases checked
260
Negative cases checked
16,143
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive260 cases, 1.59% 193true positives 67false negatives
Negative16,143 cases 1,626false positives 14,517true negatives

16,403 cases were checked, prevalence was 1.6%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 7 times more likely in a patient that has this condition than in one that does not. About 1 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 74 in every 100 patients that had this condition, and missed 67 of 260. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Enlarged heart
89.9%
92.6%
 25.5% 2.73% 0.968 35,856

Full metric set

Overall accuracy
92.5%
Positive predictive value (PPV)
25.5%
Prevalence in the cases checked
2.73%
Area under curve (AUC)
0.968
Negative predictive value (NPV)
99.7%
Positive likelihood ratio
12.2
Positive cases checked
980
Negative cases checked
34,876
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive980 cases, 2.73% 881true positives 99false negatives
Negative34,876 cases 2,576false positives 32,300true negatives

35,856 cases were checked, prevalence was 2.7%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 12 times more likely in a patient that has this condition than in one that does not. About 3 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 90 in every 100 patients that had this condition, and missed 99 of 980. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineThorax
Enlarged heart
77.0%
92.4%
 13.7% 1.54% 0.925 22,580

Full metric set

Overall accuracy
92.2%
Positive predictive value (PPV)
13.7%
Prevalence in the cases checked
1.54%
Area under curve (AUC)
0.925
Negative predictive value (NPV)
99.6%
Positive likelihood ratio
10.2
Positive cases checked
348
Negative cases checked
22,232
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive348 cases, 1.54% 268true positives 80false negatives
Negative22,232 cases 1,685false positives 20,547true negatives

22,580 cases were checked, prevalence was 1.5%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 10 times more likely in a patient that has this condition than in one that does not. About 1 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 77 in every 100 patients that had this condition, and missed 80 of 348. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Enlarged left atrium
75.3%
99.3%
 29.3% 0.36% 0.980 40,678

Full metric set

Overall accuracy
99.3%
Positive predictive value (PPV)
29.3%
Prevalence in the cases checked
0.36%
Area under curve (AUC)
0.980
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
115.2
Positive cases checked
146
Negative cases checked
40,532
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive146 cases, 0.36% 110true positives 36false negatives
Negative40,532 cases 265false positives 40,267true negatives

40,678 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is more than 100 times more likely in a patient that has this condition than in one that does not. About 3 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 110 of the 146 patients that had this condition, and missed 36. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 68 to about 82 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Enlarged left auricular appendage
85.0%
97.7%
 34.9% 1.42% 0.981 45,545

Full metric set

Overall accuracy
97.5%
Positive predictive value (PPV)
34.9%
Prevalence in the cases checked
1.42%
Area under curve (AUC)
0.981
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
37.1
Positive cases checked
647
Negative cases checked
44,898
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive647 cases, 1.42% 550true positives 97false negatives
Negative44,898 cases 1,028false positives 43,870true negatives

45,545 cases were checked, prevalence was 1.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 37 times more likely in a patient that has this condition than in one that does not. About 3 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 85 in every 100 patients that had this condition, and missed 97 of 647. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Enlarged main pulmonary artery
72.1%
90.2%
 2.9% 0.40% 0.896 36,725

Full metric set

Overall accuracy
90.2%
Positive predictive value (PPV)
2.9%
Prevalence in the cases checked
0.40%
Area under curve (AUC)
0.896
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
7.4
Positive cases checked
147
Negative cases checked
36,578
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive147 cases, 0.40% 106true positives 41false negatives
Negative36,578 cases 3,567false positives 33,011true negatives

36,725 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 7 times more likely in a patient that has this condition than in one that does not. About 3 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 106 of the 147 patients that had this condition, and missed 41. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 64 to about 79 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Enlarged pulmonary arteries
73.7%
81.3%
 6.7% 1.79% 0.865 26,071

Full metric set

Overall accuracy
81.2%
Positive predictive value (PPV)
6.7%
Prevalence in the cases checked
1.79%
Area under curve (AUC)
0.865
Negative predictive value (NPV)
99.4%
Positive likelihood ratio
3.9
Positive cases checked
467
Negative cases checked
25,604
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive467 cases, 1.79% 344true positives 123false negatives
Negative25,604 cases 4,779false positives 20,825true negatives

26,071 cases were checked, prevalence was 1.8%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 4 times more likely in a patient that has this condition than in one that does not. About 7 in every 100 positive results were confirmed, a consequence of that rarity. Anything below 5 is considered a small change, so a positive result supports the finding without confirming it.

Not noted. This classifier found about 74 in every 100 patients that had this condition, and missed 123 of 467. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Enlarged pulmonary veins
82.8%
90.1%
 7.0% 0.89% 0.934 26,784

Full metric set

Overall accuracy
90.1%
Positive predictive value (PPV)
7.0%
Prevalence in the cases checked
0.89%
Area under curve (AUC)
0.934
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
8.4
Positive cases checked
238
Negative cases checked
26,546
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive238 cases, 0.89% 197true positives 41false negatives
Negative26,546 cases 2,621false positives 23,925true negatives

26,784 cases were checked, prevalence was 0.9%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 8 times more likely in a patient that has this condition than in one that does not. About 7 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 83 in every 100 patients that had this condition, and missed 41 of 238. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineThorax
Enlarged pulmonary vessels
71.3%
90.6%
 15.0% 2.26% 0.884 8,613

Full metric set

Overall accuracy
90.2%
Positive predictive value (PPV)
15.0%
Prevalence in the cases checked
2.26%
Area under curve (AUC)
0.884
Negative predictive value (NPV)
99.3%
Positive likelihood ratio
7.6
Positive cases checked
195
Negative cases checked
8,418
Release/retrain date
Sep 2026

What the board-certified radiologist labels showed

Positive195 cases, 2.26% 139true positives 56false negatives
Negative8,418 cases 790false positives 7,628true negatives

8,613 cases were checked, prevalence was 2.3%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 8 times more likely in a patient that has this condition than in one that does not. About 1 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 139 of the 195 patients that had this condition, and missed 56. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 65 to about 77 in every 100. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Enlarged spleen
78.6%
91.5%
 6.1% 0.70% 0.936 38,917

Full metric set

Overall accuracy
91.4%
Positive predictive value (PPV)
6.1%
Prevalence in the cases checked
0.70%
Area under curve (AUC)
0.936
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
9.2
Positive cases checked
271
Negative cases checked
38,646
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive271 cases, 0.70% 213true positives 58false negatives
Negative38,646 cases 3,290false positives 35,356true negatives

38,917 cases were checked, prevalence was 0.7%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 9 times more likely in a patient that has this condition than in one that does not. About 6 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 79 in every 100 patients that had this condition, and missed 58 of 271. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Enlarged spleen
72.2%
90.9%
 5.5% 0.73% 0.887 43,842

Full metric set

Overall accuracy
90.8%
Positive predictive value (PPV)
5.5%
Prevalence in the cases checked
0.73%
Area under curve (AUC)
0.887
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
8.0
Positive cases checked
320
Negative cases checked
43,522
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive320 cases, 0.73% 231true positives 89false negatives
Negative43,522 cases 3,946false positives 39,576true negatives

43,842 cases were checked, prevalence was 0.7%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 8 times more likely in a patient that has this condition than in one that does not. About 6 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 72 in every 100 patients that had this condition, and missed 89 of 320. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineThorax
Enlarged thoracic lymph nodes
74.4%
92.5%
 7.2% 0.78% 0.887 20,635

Full metric set

Overall accuracy
92.4%
Positive predictive value (PPV)
7.2%
Prevalence in the cases checked
0.78%
Area under curve (AUC)
0.887
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
10.0
Positive cases checked
160
Negative cases checked
20,475
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive160 cases, 0.78% 119true positives 41false negatives
Negative20,475 cases 1,528false positives 18,947true negatives

20,635 cases were checked, prevalence was 0.8%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 10 times more likely in a patient that has this condition than in one that does not. About 7 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 119 of the 160 patients that had this condition, and missed 41. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 67 to about 81 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Gastric foreign material
72.7%
76.2%
 2.9% 0.95% 0.777 37,650

Full metric set

Overall accuracy
76.2%
Positive predictive value (PPV)
2.9%
Prevalence in the cases checked
0.95%
Area under curve (AUC)
0.777
Negative predictive value (NPV)
99.7%
Positive likelihood ratio
3.1
Positive cases checked
359
Negative cases checked
37,291
Release/retrain date
Mar 2026

What the board-certified radiologist labels showed

Positive359 cases, 0.95% 261true positives 98false negatives
Negative37,291 cases 8,873false positives 28,418true negatives

37,650 cases were checked, prevalence was 1.0%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 3 times more likely in a patient that has this condition than in one that does not. About 3 in every 100 positive results were confirmed, a consequence of that rarity. Anything below 5 is considered a small change, so a positive result supports the finding without confirming it.

Not noted. This classifier found about 73 in every 100 patients that had this condition, and missed 98 of 359. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Gastric malposition
74.5%
85.1%
 2.7% 0.56% 0.858 8,417

Full metric set

Overall accuracy
85.1%
Positive predictive value (PPV)
2.7%
Prevalence in the cases checked
0.56%
Area under curve (AUC)
0.858
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
5.0
Positive cases checked
47
Negative cases checked
8,370
Release/retrain date
Jun 2026

What the board-certified radiologist labels showed

Positive47 cases, 0.56% 35true positives 12false negatives
Negative8,370 cases 1,246false positives 7,124true negatives

8,417 cases were checked, prevalence was 0.6%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 5 times more likely in a patient that has this condition than in one that does not. About 3 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 35 of the 47 patients that had this condition, and missed 12. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 60 to about 85 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Heart base mass
71.2%
89.9%
 3.0% 0.44% 0.904 36,449

Full metric set

Overall accuracy
89.8%
Positive predictive value (PPV)
3.0%
Prevalence in the cases checked
0.44%
Area under curve (AUC)
0.904
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
7.1
Positive cases checked
160
Negative cases checked
36,289
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive160 cases, 0.44% 114true positives 46false negatives
Negative36,289 cases 3,654false positives 32,635true negatives

36,449 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 7 times more likely in a patient that has this condition than in one that does not. About 3 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 114 of the 160 patients that had this condition, and missed 46. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 64 to about 78 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Heart failure pattern
95.9%
92.8%
 5.2% 0.41% 0.979 35,626

Full metric set

Overall accuracy
92.9%
Positive predictive value (PPV)
5.2%
Prevalence in the cases checked
0.41%
Area under curve (AUC)
0.979
Negative predictive value (NPV)
100.0%
Positive likelihood ratio
13.4
Positive cases checked
146
Negative cases checked
35,480
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive146 cases, 0.41% 140true positives 6false negatives
Negative35,480 cases 2,541false positives 32,939true negatives

35,626 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 13 times more likely in a patient that has this condition than in one that does not. About 5 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 96 in every 100 patients that had this condition, and missed 6 of 146. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Hepatomegaly
73.6%
97.3%
 64.1% 6.08% 0.963 31,506

Full metric set

Overall accuracy
95.9%
Positive predictive value (PPV)
64.1%
Prevalence in the cases checked
6.08%
Area under curve (AUC)
0.963
Negative predictive value (NPV)
98.3%
Positive likelihood ratio
27.6
Positive cases checked
1,914
Negative cases checked
29,592
Release/retrain date
Feb 2026

What the board-certified radiologist labels showed

Positive1,914 cases, 6.08% 1,408true positives 506false negatives
Negative29,592 cases 788false positives 28,804true negatives

31,506 cases were checked, prevalence was 6.1%.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. A positive result is about 28 times more likely in a patient that has this condition than in one that does not. About 6 in every 10 positive results were confirmed.

Not noted. This classifier found about 74 in every 100 patients that had this condition, and missed 506 of 1,914. About 98 in every 100 negative results were correct. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Hepatomegaly
72.3%
93.0%
 4.6% 0.46% 0.923 29,488

Full metric set

Overall accuracy
92.9%
Positive predictive value (PPV)
4.6%
Prevalence in the cases checked
0.46%
Area under curve (AUC)
0.923
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
10.3
Positive cases checked
137
Negative cases checked
29,351
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive137 cases, 0.46% 99true positives 38false negatives
Negative29,351 cases 2,061false positives 27,290true negatives

29,488 cases were checked, prevalence was 0.5%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 10 times more likely in a patient that has this condition than in one that does not. About 5 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 99 of the 137 patients that had this condition, and missed 38. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 64 to about 79 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Hiatal hernia
75.6%
90.3%
 1.6% 0.21% 0.897 36,641

Full metric set

Overall accuracy
90.2%
Positive predictive value (PPV)
1.6%
Prevalence in the cases checked
0.21%
Area under curve (AUC)
0.897
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
7.8
Positive cases checked
78
Negative cases checked
36,563
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive78 cases, 0.21% 59true positives 19false negatives
Negative36,563 cases 3,560false positives 33,003true negatives

36,641 cases were checked, prevalence was 0.2%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 8 times more likely in a patient that has this condition than in one that does not. About 2 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 59 of the 78 patients that had this condition, and missed 19. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 65 to about 84 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

Canine & FelineSpine and MSK
Hip dysplasia
80.9%
90.9%
 34.5% 5.59% 0.934 14,665

Full metric set

Overall accuracy
90.3%
Positive predictive value (PPV)
34.5%
Prevalence in the cases checked
5.59%
Area under curve (AUC)
0.934
Negative predictive value (NPV)
98.8%
Positive likelihood ratio
8.9
Positive cases checked
820
Negative cases checked
13,845
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive820 cases, 5.59% 663true positives 157false negatives
Negative13,845 cases 1,260false positives 12,585true negatives

14,665 cases were checked, prevalence was 5.6%.

Compared with the other spine and MSK classifiers

This classifier is the highlighted mark. Gray marks are the other 10 spine and MSK classifiers.

Interpreting your AI Screening Report

Reported. A positive result is about 9 times more likely in a patient that has this condition than in one that does not. About 3 in every 10 positive results were confirmed.

Not noted. This classifier found about 81 in every 100 patients that had this condition, and missed 157 of 820. About 99 in every 100 negative results were correct. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

Canine & FelineSpine and MSK
Intervertebral disc disease, lumbar spine
76.6%
87.6%
 47.7% 12.84% 0.905 25,396

Full metric set

Overall accuracy
86.2%
Positive predictive value (PPV)
47.7%
Prevalence in the cases checked
12.84%
Area under curve (AUC)
0.905
Negative predictive value (NPV)
96.2%
Positive likelihood ratio
6.2
Positive cases checked
3,260
Negative cases checked
22,136
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive3,260 cases, 12.84% 2,497true positives 763false negatives
Negative22,136 cases 2,737false positives 19,399true negatives

25,396 cases were checked, prevalence was 13%.

Compared with the other spine and MSK classifiers

This classifier is the highlighted mark. Gray marks are the other 10 spine and MSK classifiers.

Interpreting your AI Screening Report

Reported. A positive result is about 6 times more likely in a patient that has this condition than in one that does not. About 5 in every 10 positive results were confirmed.

Not noted. This classifier found about 77 in every 100 patients that had this condition, and missed 763 of 3,260. About 96 in every 100 negative results were correct. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

Canine & FelineSpine and MSK
Intervertebral disc disease, thoracic spine
72.6%
89.6%
 2.8% 0.42% 0.904 22,683

Full metric set

Overall accuracy
89.5%
Positive predictive value (PPV)
2.8%
Prevalence in the cases checked
0.42%
Area under curve (AUC)
0.904
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
7.0
Positive cases checked
95
Negative cases checked
22,588
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive95 cases, 0.42% 69true positives 26false negatives
Negative22,588 cases 2,357false positives 20,231true negatives

22,683 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other spine and MSK classifiers

This classifier is the highlighted mark. Gray marks are the other 10 spine and MSK classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 7 times more likely in a patient that has this condition than in one that does not. About 3 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 69 of the 95 patients that had this condition, and missed 26. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 63 to about 81 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Kidney stones (left kidney)
72.7%
97.2%
 7.5% 0.31% 0.935 38,481

Full metric set

Overall accuracy
97.1%
Positive predictive value (PPV)
7.5%
Prevalence in the cases checked
0.31%
Area under curve (AUC)
0.935
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
25.7
Positive cases checked
121
Negative cases checked
38,360
Release/retrain date
Nov 2024

What the board-certified radiologist labels showed

Positive121 cases, 0.31% 88true positives 33false negatives
Negative38,360 cases 1,086false positives 37,274true negatives

38,481 cases were checked, prevalence was 0.3%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 26 times more likely in a patient that has this condition than in one that does not. About 7 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 88 of the 121 patients that had this condition, and missed 33. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 64 to about 80 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Kidney stones (left kidney)
81.7%
98.1%
 14.1% 0.38% 0.970 27,360

Full metric set

Overall accuracy
98.0%
Positive predictive value (PPV)
14.1%
Prevalence in the cases checked
0.38%
Area under curve (AUC)
0.970
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
43.1
Positive cases checked
104
Negative cases checked
27,256
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive104 cases, 0.38% 85true positives 19false negatives
Negative27,256 cases 517false positives 26,739true negatives

27,360 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 43 times more likely in a patient that has this condition than in one that does not. About 1 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 85 of the 104 patients that had this condition, and missed 19. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 73 to about 88 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Kidney stones (right kidney)
77.9%
95.2%
 4.6% 0.29% 0.918 47,930

Full metric set

Overall accuracy
95.2%
Positive predictive value (PPV)
4.6%
Prevalence in the cases checked
0.29%
Area under curve (AUC)
0.918
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
16.3
Positive cases checked
140
Negative cases checked
47,790
Release/retrain date
Nov 2024

What the board-certified radiologist labels showed

Positive140 cases, 0.29% 109true positives 31false negatives
Negative47,790 cases 2,286false positives 45,504true negatives

47,930 cases were checked, prevalence was 0.3%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 16 times more likely in a patient that has this condition than in one that does not. About 5 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 109 of the 140 patients that had this condition, and missed 31. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 70 to about 84 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Kidney stones (right kidney)
71.4%
94.6%
 18.6% 1.69% 0.889 8,281

Full metric set

Overall accuracy
94.2%
Positive predictive value (PPV)
18.6%
Prevalence in the cases checked
1.69%
Area under curve (AUC)
0.889
Negative predictive value (NPV)
99.5%
Positive likelihood ratio
13.3
Positive cases checked
140
Negative cases checked
8,141
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive140 cases, 1.69% 100true positives 40false negatives
Negative8,141 cases 437false positives 7,704true negatives

8,281 cases were checked, prevalence was 1.7%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 13 times more likely in a patient that has this condition than in one that does not. About 2 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 100 of the 140 patients that had this condition, and missed 40. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 63 to about 78 in every 100. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Left kidney margin change
73.0%
87.9%
 12.4% 2.29% 0.895 10,817

Full metric set

Overall accuracy
87.5%
Positive predictive value (PPV)
12.4%
Prevalence in the cases checked
2.29%
Area under curve (AUC)
0.895
Negative predictive value (NPV)
99.3%
Positive likelihood ratio
6.0
Positive cases checked
248
Negative cases checked
10,569
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive248 cases, 2.29% 181true positives 67false negatives
Negative10,569 cases 1,284false positives 9,285true negatives

10,817 cases were checked, prevalence was 2.3%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 6 times more likely in a patient that has this condition than in one that does not. About 1 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 73 in every 100 patients that had this condition, and missed 67 of 248. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Left kidney size change
65.2%
78.5%
 0.9% 0.31% 0.793 35,722

Full metric set

Overall accuracy
78.5%
Positive predictive value (PPV)
0.9%
Prevalence in the cases checked
0.31%
Area under curve (AUC)
0.793
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
3.0
Positive cases checked
112
Negative cases checked
35,610
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive112 cases, 0.31% 73true positives 39false negatives
Negative35,610 cases 7,658false positives 27,952true negatives

35,722 cases were checked, prevalence was 0.3%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 3 times more likely in a patient that has this condition than in one that does not. About 9 in every 1,000 positive results were confirmed, a consequence of that rarity. Anything below 5 is considered a small change, so a positive result supports the finding without confirming it.

Not noted. This classifier found 73 of the 112 patients that had this condition, and missed 39. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 56 to about 73 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Left kidney size change
71.7%
83.8%
 2.6% 0.59% 0.861 24,686

Full metric set

Overall accuracy
83.7%
Positive predictive value (PPV)
2.6%
Prevalence in the cases checked
0.59%
Area under curve (AUC)
0.861
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
4.4
Positive cases checked
145
Negative cases checked
24,541
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive145 cases, 0.59% 104true positives 41false negatives
Negative24,541 cases 3,971false positives 20,570true negatives

24,686 cases were checked, prevalence was 0.6%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 4 times more likely in a patient that has this condition than in one that does not. About 3 in every 100 positive results were confirmed, a consequence of that rarity. Anything below 5 is considered a small change, so a positive result supports the finding without confirming it.

Not noted. This classifier found 104 of the 145 patients that had this condition, and missed 41. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 64 to about 78 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Left mediastinal shift
74.2%
74.4%
 1.9% 0.68% Not published 43,910

Full metric set

Overall accuracy
74.4%
Positive predictive value (PPV)
1.9%
Prevalence in the cases checked
0.68%
Area under curve (AUC)
Not published
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
2.9
Positive cases checked
298
Negative cases checked
43,612
Release/retrain date
Dec 2025

What the board-certified radiologist labels showed

Positive298 cases, 0.68% 221true positives 77false negatives
Negative43,612 cases 11,160false positives 32,452true negatives

43,910 cases were checked, prevalence was 0.7%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 2.9 times more likely in a patient that has this condition than in one that does not. About 2 in every 100 positive results were confirmed, a consequence of that rarity. Anything below 5 is considered a small change, so a positive result supports the finding without confirming it.

Not noted. This classifier found about 74 in every 100 patients that had this condition, and missed 77 of 298. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Left-sided cardiomegaly
96.3%
98.2%
 6.1% 0.12% 0.995 45,382

Full metric set

Overall accuracy
98.2%
Positive predictive value (PPV)
6.1%
Prevalence in the cases checked
0.12%
Area under curve (AUC)
0.995
Negative predictive value (NPV)
100.0%
Positive likelihood ratio
54.2
Positive cases checked
54
Negative cases checked
45,328
Release/retrain date
May 2026

What the board-certified radiologist labels showed

Positive54 cases, 0.12% 52true positives 2false negatives
Negative45,328 cases 806false positives 44,522true negatives

45,382 cases were checked, prevalence was 0.1%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 54 times more likely in a patient that has this condition than in one that does not. About 6 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 96 in every 100 patients that had this condition, and missed 2 of 54. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineThorax
Left-sided cardiomegaly
70.2%
92.9%
 25.2% 3.29% 0.902 8,263

Full metric set

Overall accuracy
92.2%
Positive predictive value (PPV)
25.2%
Prevalence in the cases checked
3.29%
Area under curve (AUC)
0.902
Negative predictive value (NPV)
98.9%
Positive likelihood ratio
9.9
Positive cases checked
272
Negative cases checked
7,991
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive272 cases, 3.29% 191true positives 81false negatives
Negative7,991 cases 567false positives 7,424true negatives

8,263 cases were checked, prevalence was 3.3%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 10 times more likely in a patient that has this condition than in one that does not. About 3 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 70 in every 100 patients that had this condition, and missed 81 of 272. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineThorax
Left-sided heart failure pattern
92.9%
94.7%
 7.0% 0.43% 0.974 33,080

Full metric set

Overall accuracy
94.7%
Positive predictive value (PPV)
7.0%
Prevalence in the cases checked
0.43%
Area under curve (AUC)
0.974
Negative predictive value (NPV)
100.0%
Positive likelihood ratio
17.5
Positive cases checked
141
Negative cases checked
32,939
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive141 cases, 0.43% 131true positives 10false negatives
Negative32,939 cases 1,747false positives 31,192true negatives

33,080 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 18 times more likely in a patient that has this condition than in one that does not. About 7 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 93 in every 100 patients that had this condition, and missed 10 of 141. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Liver mass
74.0%
94.1%
 4.7% 0.39% 0.942 31,504

Full metric set

Overall accuracy
94.1%
Positive predictive value (PPV)
4.7%
Prevalence in the cases checked
0.39%
Area under curve (AUC)
0.942
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
12.6
Positive cases checked
123
Negative cases checked
31,381
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive123 cases, 0.39% 91true positives 32false negatives
Negative31,381 cases 1,840false positives 29,541true negatives

31,504 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 13 times more likely in a patient that has this condition than in one that does not. About 5 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 91 of the 123 patients that had this condition, and missed 32. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 66 to about 81 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Liver mass
75.5%
92.6%
 3.2% 0.32% 0.910 45,926

Full metric set

Overall accuracy
92.6%
Positive predictive value (PPV)
3.2%
Prevalence in the cases checked
0.32%
Area under curve (AUC)
0.910
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
10.3
Positive cases checked
147
Negative cases checked
45,779
Release/retrain date
Sep 2026

What the board-certified radiologist labels showed

Positive147 cases, 0.32% 111true positives 36false negatives
Negative45,779 cases 3,371false positives 42,408true negatives

45,926 cases were checked, prevalence was 0.3%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 10 times more likely in a patient that has this condition than in one that does not. About 3 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 111 of the 147 patients that had this condition, and missed 36. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 68 to about 82 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Microcardia
81.1%
85.3%
 1.0% 0.19% 0.904 48,242

Full metric set

Overall accuracy
85.3%
Positive predictive value (PPV)
1.0%
Prevalence in the cases checked
0.19%
Area under curve (AUC)
0.904
Negative predictive value (NPV)
100.0%
Positive likelihood ratio
5.5
Positive cases checked
90
Negative cases checked
48,152
Release/retrain date
Mar 2026

What the board-certified radiologist labels showed

Positive90 cases, 0.19% 73true positives 17false negatives
Negative48,152 cases 7,071false positives 41,081true negatives

48,242 cases were checked, prevalence was 0.2%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 6 times more likely in a patient that has this condition than in one that does not. About 1 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 73 of the 90 patients that had this condition, and missed 17. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 72 to about 88 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Mid-abdominal mass
81.7%
73.4%
 2.9% 0.97% Not published 13,493

Full metric set

Overall accuracy
73.5%
Positive predictive value (PPV)
2.9%
Prevalence in the cases checked
0.97%
Area under curve (AUC)
Not published
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
3.1
Positive cases checked
131
Negative cases checked
13,362
Release/retrain date
Dec 2025

What the board-certified radiologist labels showed

Positive131 cases, 0.97% 107true positives 24false negatives
Negative13,362 cases 3,550false positives 9,812true negatives

13,493 cases were checked, prevalence was 1.0%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 3 times more likely in a patient that has this condition than in one that does not. About 3 in every 100 positive results were confirmed, a consequence of that rarity. Anything below 5 is considered a small change, so a positive result supports the finding without confirming it.

Not noted. This classifier found 107 of the 131 patients that had this condition, and missed 24. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 74 to about 87 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Miliary pulmonary pattern
79.7%
97.3%
 23.2% 1.00% 0.957 34,735

Full metric set

Overall accuracy
97.1%
Positive predictive value (PPV)
23.2%
Prevalence in the cases checked
1.00%
Area under curve (AUC)
0.957
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
29.8
Positive cases checked
349
Negative cases checked
34,386
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive349 cases, 1.00% 278true positives 71false negatives
Negative34,386 cases 919false positives 33,467true negatives

34,735 cases were checked, prevalence was 1.0%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 30 times more likely in a patient that has this condition than in one that does not. About 2 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 80 in every 100 patients that had this condition, and missed 71 of 349. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Mineral in the gallbladder or biliary tract
76.1%
94.8%
 15.1% 1.20% 0.926 46,512

Full metric set

Overall accuracy
94.6%
Positive predictive value (PPV)
15.1%
Prevalence in the cases checked
1.20%
Area under curve (AUC)
0.926
Negative predictive value (NPV)
99.7%
Positive likelihood ratio
14.7
Positive cases checked
557
Negative cases checked
45,955
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive557 cases, 1.20% 424true positives 133false negatives
Negative45,955 cases 2,386false positives 43,569true negatives

46,512 cases were checked, prevalence was 1.2%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 15 times more likely in a patient that has this condition than in one that does not. About 2 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 76 in every 100 patients that had this condition, and missed 133 of 557. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Mineral or metal opaque gastric material
71.9%
91.3%
 14.6% 2.03% 0.875 22,086

Full metric set

Overall accuracy
90.9%
Positive predictive value (PPV)
14.6%
Prevalence in the cases checked
2.03%
Area under curve (AUC)
0.875
Negative predictive value (NPV)
99.4%
Positive likelihood ratio
8.3
Positive cases checked
448
Negative cases checked
21,638
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive448 cases, 2.03% 322true positives 126false negatives
Negative21,638 cases 1,881false positives 19,757true negatives

22,086 cases were checked, prevalence was 2.0%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 8 times more likely in a patient that has this condition than in one that does not. About 1 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 72 in every 100 patients that had this condition, and missed 126 of 448. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Mineral or metal opaque small intestinal material
69.7%
90.7%
 9.6% 1.40% 0.879 37,512

Full metric set

Overall accuracy
90.4%
Positive predictive value (PPV)
9.6%
Prevalence in the cases checked
1.40%
Area under curve (AUC)
0.879
Negative predictive value (NPV)
99.5%
Positive likelihood ratio
7.5
Positive cases checked
524
Negative cases checked
36,988
Release/retrain date
Sep 2026

What the board-certified radiologist labels showed

Positive524 cases, 1.40% 365true positives 159false negatives
Negative36,988 cases 3,426false positives 33,562true negatives

37,512 cases were checked, prevalence was 1.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 8 times more likely in a patient that has this condition than in one that does not. About 1 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 70 in every 100 patients that had this condition, and missed 159 of 524. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Narrowed intrathoracic trachea
69.6%
90.5%
 22.9% 3.91% 0.904 35,694

Full metric set

Overall accuracy
89.7%
Positive predictive value (PPV)
22.9%
Prevalence in the cases checked
3.91%
Area under curve (AUC)
0.904
Negative predictive value (NPV)
98.7%
Positive likelihood ratio
7.3
Positive cases checked
1,396
Negative cases checked
34,298
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive1,396 cases, 3.91% 972true positives 424false negatives
Negative34,298 cases 3,267false positives 31,031true negatives

35,694 cases were checked, prevalence was 3.9%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 7 times more likely in a patient that has this condition than in one that does not. About 2 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 70 in every 100 patients that had this condition, and missed 424 of 1,396. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Narrowed or compressed bronchi
89.9%
90.9%
 3.9% 0.41% 0.962 36,396

Full metric set

Overall accuracy
90.9%
Positive predictive value (PPV)
3.9%
Prevalence in the cases checked
0.41%
Area under curve (AUC)
0.962
Negative predictive value (NPV)
100.0%
Positive likelihood ratio
9.9
Positive cases checked
149
Negative cases checked
36,247
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive149 cases, 0.41% 134true positives 15false negatives
Negative36,247 cases 3,295false positives 32,952true negatives

36,396 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 10 times more likely in a patient that has this condition than in one that does not. About 4 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 90 in every 100 patients that had this condition, and missed 15 of 149. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Obscuring pleural effusion
76.6%
99.2%
 67.2% 2.01% 0.965 25,721

Full metric set

Overall accuracy
98.8%
Positive predictive value (PPV)
67.2%
Prevalence in the cases checked
2.01%
Area under curve (AUC)
0.965
Negative predictive value (NPV)
99.5%
Positive likelihood ratio
99.6
Positive cases checked
518
Negative cases checked
25,203
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive518 cases, 2.01% 397true positives 121false negatives
Negative25,203 cases 194false positives 25,009true negatives

25,721 cases were checked, prevalence was 2.0%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 100 times more likely in a patient that has this condition than in one that does not. About 7 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 77 in every 100 patients that had this condition, and missed 121 of 518. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineThorax
Obscuring pleural effusion
86.2%
96.2%
 32.0% 2.01% 0.969 25,214

Full metric set

Overall accuracy
96.0%
Positive predictive value (PPV)
32.0%
Prevalence in the cases checked
2.01%
Area under curve (AUC)
0.969
Negative predictive value (NPV)
99.7%
Positive likelihood ratio
22.9
Positive cases checked
506
Negative cases checked
24,708
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive506 cases, 2.01% 436true positives 70false negatives
Negative24,708 cases 928false positives 23,780true negatives

25,214 cases were checked, prevalence was 2.0%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 23 times more likely in a patient that has this condition than in one that does not. About 3 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 86 in every 100 patients that had this condition, and missed 70 of 506. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

Canine & FelineSpine and MSK
Periarticular proliferation, pelvis
85.0%
95.3%
 89.8% 32.77% 0.963 6,452

Full metric set

Overall accuracy
91.9%
Positive predictive value (PPV)
89.8%
Prevalence in the cases checked
32.77%
Area under curve (AUC)
0.963
Negative predictive value (NPV)
92.9%
Positive likelihood ratio
18.0
Positive cases checked
2,114
Negative cases checked
4,338
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive2,114 cases, 32.77% 1,797true positives 317false negatives
Negative4,338 cases 205false positives 4,133true negatives

6,452 cases were checked, prevalence was 33%.

Compared with the other spine and MSK classifiers

This classifier is the highlighted mark. Gray marks are the other 10 spine and MSK classifiers.

Interpreting your AI Screening Report

Reported. A positive result is about 18 times more likely in a patient that has this condition than in one that does not. About 9 in every 10 positive results were confirmed.

Not noted. This classifier found about 85 in every 100 patients that had this condition, and missed 317 of 2,114. About 93 in every 100 negative results were correct. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Perihilar infiltrate
91.3%
90.9%
 4.0% 0.42% 0.956 35,578

Full metric set

Overall accuracy
90.9%
Positive predictive value (PPV)
4.0%
Prevalence in the cases checked
0.42%
Area under curve (AUC)
0.956
Negative predictive value (NPV)
100.0%
Positive likelihood ratio
10.0
Positive cases checked
149
Negative cases checked
35,429
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive149 cases, 0.42% 136true positives 13false negatives
Negative35,429 cases 3,229false positives 32,200true negatives

35,578 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 10 times more likely in a patient that has this condition than in one that does not. About 4 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 91 in every 100 patients that had this condition, and missed 13 of 149. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Pleural effusion
90.9%
99.6%
 11.4% 0.06% 0.994 36,552

Full metric set

Overall accuracy
99.6%
Positive predictive value (PPV)
11.4%
Prevalence in the cases checked
0.06%
Area under curve (AUC)
0.994
Negative predictive value (NPV)
100.0%
Positive likelihood ratio
214.3
Positive cases checked
22
Negative cases checked
36,530
Release/retrain date
May 2026

What the board-certified radiologist labels showed

Positive22 cases, 0.06% 20true positives 2false negatives
Negative36,530 cases 155false positives 36,375true negatives

36,552 cases were checked, prevalence was 0.1%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is more than 100 times more likely in a patient that has this condition than in one that does not. About 1 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 20 of the 22 patients that had this condition, and missed 2. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 72 to about 97 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineThorax
Pleural effusion
88.1%
99.3%
 68.8% 1.63% 0.990 35,131

Full metric set

Overall accuracy
99.2%
Positive predictive value (PPV)
68.8%
Prevalence in the cases checked
1.63%
Area under curve (AUC)
0.990
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
133.0
Positive cases checked
573
Negative cases checked
34,558
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive573 cases, 1.63% 505true positives 68false negatives
Negative34,558 cases 229false positives 34,329true negatives

35,131 cases were checked, prevalence was 1.6%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is more than 100 times more likely in a patient that has this condition than in one that does not. About 7 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 88 in every 100 patients that had this condition, and missed 68 of 573. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Pregnancy (mineralized fetal skeletons)
99.1%
97.8%
 12.4% 0.31% 0.998 37,726

Full metric set

Overall accuracy
97.8%
Positive predictive value (PPV)
12.4%
Prevalence in the cases checked
0.31%
Area under curve (AUC)
0.998
Negative predictive value (NPV)
100.0%
Positive likelihood ratio
45.3
Positive cases checked
117
Negative cases checked
37,609
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive117 cases, 0.31% 116true positives 1false negatives
Negative37,609 cases 823false positives 36,786true negatives

37,726 cases were checked, prevalence was 0.3%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 45 times more likely in a patient that has this condition than in one that does not. About 1 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 99 in every 100 patients that had this condition, and missed 1 of 117. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Pregnancy (mineralized fetal skeletons)
100.0%
99.4%
 25.3% 0.21% 0.999 9,879

Full metric set

Overall accuracy
99.4%
Positive predictive value (PPV)
25.3%
Prevalence in the cases checked
0.21%
Area under curve (AUC)
0.999
Negative predictive value (NPV)
100.0%
Positive likelihood ratio
159.0
Positive cases checked
21
Negative cases checked
9,858
Release/retrain date
Oct 2025

What the board-certified radiologist labels showed

Positive21 cases, 0.21% 21true positives 0false negatives
Negative9,858 cases 62false positives 9,796true negatives

9,879 cases were checked, prevalence was 0.2%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is more than 100 times more likely in a patient that has this condition than in one that does not. About 3 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 21 of the 21 patients that had this condition, and missed 0. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 85 to about 100 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Pulmonary edema
91.0%
95.1%
 23.5% 1.64% 0.982 8,086

Full metric set

Overall accuracy
95.0%
Positive predictive value (PPV)
23.5%
Prevalence in the cases checked
1.64%
Area under curve (AUC)
0.982
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
18.4
Positive cases checked
133
Negative cases checked
7,953
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive133 cases, 1.64% 121true positives 12false negatives
Negative7,953 cases 393false positives 7,560true negatives

8,086 cases were checked, prevalence was 1.6%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 18 times more likely in a patient that has this condition than in one that does not. About 2 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 91 in every 100 patients that had this condition, and missed 12 of 133. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineThorax
Pulmonary edema
77.8%
96.4%
 7.4% 0.37% 0.968 34,197

Full metric set

Overall accuracy
96.4%
Positive predictive value (PPV)
7.4%
Prevalence in the cases checked
0.37%
Area under curve (AUC)
0.968
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
21.8
Positive cases checked
126
Negative cases checked
34,071
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive126 cases, 0.37% 98true positives 28false negatives
Negative34,071 cases 1,218false positives 32,853true negatives

34,197 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 22 times more likely in a patient that has this condition than in one that does not. About 7 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 98 of the 126 patients that had this condition, and missed 28. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 70 to about 84 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Pulmonary masses
73.7%
98.4%
 49.4% 2.07% 0.972 44,023

Full metric set

Overall accuracy
97.9%
Positive predictive value (PPV)
49.4%
Prevalence in the cases checked
2.07%
Area under curve (AUC)
0.972
Negative predictive value (NPV)
99.4%
Positive likelihood ratio
46.1
Positive cases checked
912
Negative cases checked
43,111
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive912 cases, 2.07% 672true positives 240false negatives
Negative43,111 cases 689false positives 42,422true negatives

44,023 cases were checked, prevalence was 2.1%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 46 times more likely in a patient that has this condition than in one that does not. About 5 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 74 in every 100 patients that had this condition, and missed 240 of 912. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineThorax
Pulmonary masses
84.5%
94.7%
 9.2% 0.63% 0.953 22,412

Full metric set

Overall accuracy
94.6%
Positive predictive value (PPV)
9.2%
Prevalence in the cases checked
0.63%
Area under curve (AUC)
0.953
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
15.9
Positive cases checked
142
Negative cases checked
22,270
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive142 cases, 0.63% 120true positives 22false negatives
Negative22,270 cases 1,183false positives 21,087true negatives

22,412 cases were checked, prevalence was 0.6%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 16 times more likely in a patient that has this condition than in one that does not. About 9 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 85 in every 100 patients that had this condition, and missed 22 of 142. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Right kidney margin change
69.1%
82.0%
 8.0% 2.23% 0.846 9,163

Full metric set

Overall accuracy
81.7%
Positive predictive value (PPV)
8.0%
Prevalence in the cases checked
2.23%
Area under curve (AUC)
0.846
Negative predictive value (NPV)
99.1%
Positive likelihood ratio
3.8
Positive cases checked
204
Negative cases checked
8,959
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive204 cases, 2.23% 141true positives 63false negatives
Negative8,959 cases 1,616false positives 7,343true negatives

9,163 cases were checked, prevalence was 2.2%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 4 times more likely in a patient that has this condition than in one that does not. About 8 in every 100 positive results were confirmed, a consequence of that rarity. Anything below 5 is considered a small change, so a positive result supports the finding without confirming it.

Not noted. This classifier found 141 of the 204 patients that had this condition, and missed 63. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 62 to about 75 in every 100. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Right kidney size change
69.0%
85.9%
 1.9% 0.40% 0.864 31,878

Full metric set

Overall accuracy
85.8%
Positive predictive value (PPV)
1.9%
Prevalence in the cases checked
0.40%
Area under curve (AUC)
0.864
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
4.9
Positive cases checked
129
Negative cases checked
31,749
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive129 cases, 0.40% 89true positives 40false negatives
Negative31,749 cases 4,476false positives 27,273true negatives

31,878 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 5 times more likely in a patient that has this condition than in one that does not. About 2 in every 100 positive results were confirmed, a consequence of that rarity. Anything below 5 is considered a small change, so a positive result supports the finding without confirming it.

Not noted. This classifier found 89 of the 129 patients that had this condition, and missed 40. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 61 to about 76 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Right kidney size change
76.2%
84.5%
 4.0% 0.83% 0.876 17,719

Full metric set

Overall accuracy
84.4%
Positive predictive value (PPV)
4.0%
Prevalence in the cases checked
0.83%
Area under curve (AUC)
0.876
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
4.9
Positive cases checked
147
Negative cases checked
17,572
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive147 cases, 0.83% 112true positives 35false negatives
Negative17,572 cases 2,721false positives 14,851true negatives

17,719 cases were checked, prevalence was 0.8%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 5 times more likely in a patient that has this condition than in one that does not. About 4 in every 100 positive results were confirmed, a consequence of that rarity. Anything below 5 is considered a small change, so a positive result supports the finding without confirming it.

Not noted. This classifier found 112 of the 147 patients that had this condition, and missed 35. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 69 to about 82 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Segmental small intestinal distension
71.2%
90.2%
 19.3% 3.19% 0.886 47,340

Full metric set

Overall accuracy
89.6%
Positive predictive value (PPV)
19.3%
Prevalence in the cases checked
3.19%
Area under curve (AUC)
0.886
Negative predictive value (NPV)
99.0%
Positive likelihood ratio
7.3
Positive cases checked
1,512
Negative cases checked
45,828
Release/retrain date
Sep 2026

What the board-certified radiologist labels showed

Positive1,512 cases, 3.19% 1,077true positives 435false negatives
Negative45,828 cases 4,501false positives 41,327true negatives

47,340 cases were checked, prevalence was 3.2%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 7 times more likely in a patient that has this condition than in one that does not. About 2 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 71 in every 100 patients that had this condition, and missed 435 of 1,512. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Segmental small intestinal distension
71.3%
90.0%
 2.3% 0.33% 0.889 45,940

Full metric set

Overall accuracy
89.9%
Positive predictive value (PPV)
2.3%
Prevalence in the cases checked
0.33%
Area under curve (AUC)
0.889
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
7.1
Positive cases checked
150
Negative cases checked
45,790
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive150 cases, 0.33% 107true positives 43false negatives
Negative45,790 cases 4,585false positives 41,205true negatives

45,940 cases were checked, prevalence was 0.3%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 7 times more likely in a patient that has this condition than in one that does not. About 2 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 107 of the 150 patients that had this condition, and missed 43. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 64 to about 78 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Small intestinal foreign material
72.6%
84.7%
 3.1% 0.67% 0.868 27,913

Full metric set

Overall accuracy
84.6%
Positive predictive value (PPV)
3.1%
Prevalence in the cases checked
0.67%
Area under curve (AUC)
0.868
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
4.7
Positive cases checked
186
Negative cases checked
27,727
Release/retrain date
Mar 2026

What the board-certified radiologist labels showed

Positive186 cases, 0.67% 135true positives 51false negatives
Negative27,727 cases 4,254false positives 23,473true negatives

27,913 cases were checked, prevalence was 0.7%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 5 times more likely in a patient that has this condition than in one that does not. About 3 in every 100 positive results were confirmed, a consequence of that rarity. Anything below 5 is considered a small change, so a positive result supports the finding without confirming it.

Not noted. This classifier found 135 of the 186 patients that had this condition, and missed 51. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 66 to about 78 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Small liver
68.6%
90.2%
 1.3% 0.19% 0.905 36,348

Full metric set

Overall accuracy
90.2%
Positive predictive value (PPV)
1.3%
Prevalence in the cases checked
0.19%
Area under curve (AUC)
0.905
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
7.0
Positive cases checked
70
Negative cases checked
36,278
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive70 cases, 0.19% 48true positives 22false negatives
Negative36,278 cases 3,556false positives 32,722true negatives

36,348 cases were checked, prevalence was 0.2%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 7 times more likely in a patient that has this condition than in one that does not. About 1 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 48 of the 70 patients that had this condition, and missed 22. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 57 to about 78 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Splenic mass
70.5%
93.3%
 16.2% 1.81% 0.909 33,798

Full metric set

Overall accuracy
92.9%
Positive predictive value (PPV)
16.2%
Prevalence in the cases checked
1.81%
Area under curve (AUC)
0.909
Negative predictive value (NPV)
99.4%
Positive likelihood ratio
10.5
Positive cases checked
611
Negative cases checked
33,187
Release/retrain date
Nov 2024

What the board-certified radiologist labels showed

Positive611 cases, 1.81% 431true positives 180false negatives
Negative33,187 cases 2,225false positives 30,962true negatives

33,798 cases were checked, prevalence was 1.8%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 11 times more likely in a patient that has this condition than in one that does not. About 2 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 71 in every 100 patients that had this condition, and missed 180 of 611. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

Canine & FelineSpine and MSK
Spondylosis deformans, lumbar spine
78.5%
93.5%
 55.3% 9.24% 0.951 16,773

Full metric set

Overall accuracy
92.1%
Positive predictive value (PPV)
55.3%
Prevalence in the cases checked
9.24%
Area under curve (AUC)
0.951
Negative predictive value (NPV)
97.7%
Positive likelihood ratio
12.1
Positive cases checked
1,550
Negative cases checked
15,223
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive1,550 cases, 9.24% 1,216true positives 334false negatives
Negative15,223 cases 983false positives 14,240true negatives

16,773 cases were checked, prevalence was 9.2%.

Compared with the other spine and MSK classifiers

This classifier is the highlighted mark. Gray marks are the other 10 spine and MSK classifiers.

Interpreting your AI Screening Report

Reported. A positive result is about 12 times more likely in a patient that has this condition than in one that does not. About 6 in every 10 positive results were confirmed.

Not noted. This classifier found about 78 in every 100 patients that had this condition, and missed 334 of 1,550. About 98 in every 100 negative results were correct. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

Canine & FelineSpine and MSK
Spondylosis deformans, lumbosacral junction
78.7%
94.3%
 4.7% 0.35% 0.957 38,674

Full metric set

Overall accuracy
94.3%
Positive predictive value (PPV)
4.7%
Prevalence in the cases checked
0.35%
Area under curve (AUC)
0.957
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
13.8
Positive cases checked
136
Negative cases checked
38,538
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive136 cases, 0.35% 107true positives 29false negatives
Negative38,538 cases 2,190false positives 36,348true negatives

38,674 cases were checked, prevalence was 0.4%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other spine and MSK classifiers

This classifier is the highlighted mark. Gray marks are the other 10 spine and MSK classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 14 times more likely in a patient that has this condition than in one that does not. About 5 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 107 of the 136 patients that had this condition, and missed 29. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 71 to about 85 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

Canine & FelineSpine and MSK
Spondylosis deformans, thoracic spine
79.4%
91.0%
 18.9% 2.59% 0.929 25,708

Full metric set

Overall accuracy
90.7%
Positive predictive value (PPV)
18.9%
Prevalence in the cases checked
2.59%
Area under curve (AUC)
0.929
Negative predictive value (NPV)
99.4%
Positive likelihood ratio
8.8
Positive cases checked
665
Negative cases checked
25,043
Release/retrain date
Sep 2026

What the board-certified radiologist labels showed

Positive665 cases, 2.59% 528true positives 137false negatives
Negative25,043 cases 2,262false positives 22,781true negatives

25,708 cases were checked, prevalence was 2.6%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other spine and MSK classifiers

This classifier is the highlighted mark. Gray marks are the other 10 spine and MSK classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 9 times more likely in a patient that has this condition than in one that does not. About 2 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 79 in every 100 patients that had this condition, and missed 137 of 665. About 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

Canine & FelineSpine and MSK
Spondylosis deformans, thoracolumbar junction
90.0%
94.5%
 10.8% 0.73% 0.967 28,653

Full metric set

Overall accuracy
94.5%
Positive predictive value (PPV)
10.8%
Prevalence in the cases checked
0.73%
Area under curve (AUC)
0.967
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
16.4
Positive cases checked
210
Negative cases checked
28,443
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive210 cases, 0.73% 189true positives 21false negatives
Negative28,443 cases 1,562false positives 26,881true negatives

28,653 cases were checked, prevalence was 0.7%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other spine and MSK classifiers

This classifier is the highlighted mark. Gray marks are the other 10 spine and MSK classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 16 times more likely in a patient that has this condition than in one that does not. About 1 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 90 in every 100 patients that had this condition, and missed 21 of 210. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Thoracic esophageal dilation
83.5%
96.0%
 25.2% 1.59% 0.961 28,548

Full metric set

Overall accuracy
95.8%
Positive predictive value (PPV)
25.2%
Prevalence in the cases checked
1.59%
Area under curve (AUC)
0.961
Negative predictive value (NPV)
99.7%
Positive likelihood ratio
20.8
Positive cases checked
454
Negative cases checked
28,094
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive454 cases, 1.59% 379true positives 75false negatives
Negative28,094 cases 1,126false positives 26,968true negatives

28,548 cases were checked, prevalence was 1.6%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 21 times more likely in a patient that has this condition than in one that does not. About 3 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 83 in every 100 patients that had this condition, and missed 75 of 454. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineThorax
Thoracic esophageal dilation
74.1%
92.7%
 3.2% 0.32% 0.892 33,972

Full metric set

Overall accuracy
92.7%
Positive predictive value (PPV)
3.2%
Prevalence in the cases checked
0.32%
Area under curve (AUC)
0.892
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
10.2
Positive cases checked
108
Negative cases checked
33,864
Release/retrain date
Aug 2026

What the board-certified radiologist labels showed

Positive108 cases, 0.32% 80true positives 28false negatives
Negative33,864 cases 2,457false positives 31,407true negatives

33,972 cases were checked, prevalence was 0.3%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 10 times more likely in a patient that has this condition than in one that does not. About 3 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 80 of the 108 patients that had this condition, and missed 28. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 65 to about 81 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Thoracic lymphadenopathy
71.4%
97.0%
 15.6% 0.78% 0.924 7,197

Full metric set

Overall accuracy
96.8%
Positive predictive value (PPV)
15.6%
Prevalence in the cases checked
0.78%
Area under curve (AUC)
0.924
Negative predictive value (NPV)
99.8%
Positive likelihood ratio
23.5
Positive cases checked
56
Negative cases checked
7,141
Release/retrain date
Jul 2026

What the board-certified radiologist labels showed

Positive56 cases, 0.78% 40true positives 16false negatives
Negative7,141 cases 217false positives 6,924true negatives

7,197 cases were checked, prevalence was 0.8%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 24 times more likely in a patient that has this condition than in one that does not. About 2 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 40 of the 56 patients that had this condition, and missed 16. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 59 to about 82 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineThorax
Underinflated lungs
71.9%
89.4%
 2.3% 0.35% 0.894 41,812

Full metric set

Overall accuracy
89.4%
Positive predictive value (PPV)
2.3%
Prevalence in the cases checked
0.35%
Area under curve (AUC)
0.894
Negative predictive value (NPV)
99.9%
Positive likelihood ratio
6.8
Positive cases checked
146
Negative cases checked
41,666
Release/retrain date
Sep 2026

What the board-certified radiologist labels showed

Positive146 cases, 0.35% 105true positives 41false negatives
Negative41,666 cases 4,396false positives 37,270true negatives

41,812 cases were checked, prevalence was 0.3%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other thorax classifiers

This classifier is the highlighted mark. Gray marks are the other 43 thorax classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 7 times more likely in a patient that has this condition than in one that does not. About 2 in every 100 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found 105 of the 146 patients that had this condition, and missed 41. That is too few cases to settle the percentage, which could reasonably sit anywhere from about 64 to about 79 in every 100. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

CanineAbdomen
Urocystoliths / urethroliths
73.7%
95.1%
 20.6% 1.71% 0.918 56,820

Full metric set

Overall accuracy
94.7%
Positive predictive value (PPV)
20.6%
Prevalence in the cases checked
1.71%
Area under curve (AUC)
0.918
Negative predictive value (NPV)
99.5%
Positive likelihood ratio
15.0
Positive cases checked
970
Negative cases checked
55,850
Release/retrain date
Sep 2026

What the board-certified radiologist labels showed

Positive970 cases, 1.71% 715true positives 255false negatives
Negative55,850 cases 2,753false positives 53,097true negatives

56,820 cases were checked, prevalence was 1.7%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 15 times more likely in a patient that has this condition than in one that does not. About 2 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 74 in every 100 patients that had this condition, and missed 255 of 970. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.

FelineAbdomen
Urocystoliths / urethroliths
71.2%
92.1%
 12.2% 1.53% 0.886 40,513

Full metric set

Overall accuracy
91.7%
Positive predictive value (PPV)
12.2%
Prevalence in the cases checked
1.53%
Area under curve (AUC)
0.886
Negative predictive value (NPV)
99.5%
Positive likelihood ratio
9.0
Positive cases checked
618
Negative cases checked
39,895
Release/retrain date
Sep 2026

What the board-certified radiologist labels showed

Positive618 cases, 1.53% 440true positives 178false negatives
Negative39,895 cases 3,170false positives 36,725true negatives

40,513 cases were checked, prevalence was 1.5%. When a condition is less prevalent in our dataset, there is a higher chance that a positive result is a false alarm, even from a classifier that is working well. We recommend confirming a positive against the images. Low prevalence can mean the condition is genuinely uncommon in practice, or that we hold fewer images of it.

Compared with the other abdomen classifiers

This classifier is the highlighted mark. Gray marks are the other 38 abdomen classifiers.

Interpreting your AI Screening Report

Reported. This is an uncommon condition in our dataset. A positive result is about 9 times more likely in a patient that has this condition than in one that does not. About 1 in every 10 positive results were confirmed, a consequence of that rarity.

Not noted. This classifier found about 71 in every 100 patients that had this condition, and missed 178 of 618. More than 99 in every 100 negative results were correct, which reflects how uncommon this condition is. On a patient you already suspect, treat this as one piece of evidence rather than a rule-out.
No classifiers match that search.Try a broader word such as heart, lung, kidney, spine, or clear the filters above.

Ready to Experience AI-Assisted Radiology?

See how Vetology’s classifiers can improve diagnostic confidence and workflow efficiency in your practice.

Request a Demo
View Pricing

Pin It on Pinterest