Model card
How Cathra works
What it was trained on, what its scores mean, and where it falls short.
What it does
You give Cathra an existing ovarian ultrasound image. It compares the image to two visual patterns it learned during training, normal ovarian appearance and abnormal ovarian appearance, and returns a similarity score for each. It does not diagnose anything.
Training data
The model was trained on a labeled set of ovarian ultrasound images, sorted into two categories:
- Normal: ovaries with a typical appearance.
- Abnormal: a broad category that includes non-cancerous findings such as polycystic ovarian appearance, simple cysts, and complex cysts. An abnormal-pattern result does not mean cancer.
Dataset credit
Training images come from the PCOSGen dataset (train and test sets), created for the Auto-PCOS classification challenge by Palak Handa, Anushka Saini, Siddhant Dutta, Harsh Pathak, Nishi Choudhary, Nidhi Goel, and Jasdeep Kaur Dhanao. Published on Zenodo, 2025: train · test. Used under the CC BY 4.0 license. Images were sorted into Normal and Abnormal using the dataset’s own labels.
Reading the results
- Each percentage is a visual similarity score: how closely the image resembles that pattern. It is not the probability that a person has any condition.
- The higher of the two scores is shown as the result label.
- If the higher score is below 60%, or the two scores are within 15 points of each other, Cathra shows “Unable to assess confidently” instead of picking one.
Limitations
- The training set is small and has not been clinically validated.
- Results on images from other hospitals, ultrasound machines, or patient populations are unknown.
- Image quality, cropping, on-screen labels, and annotations can change the scores.
- It only knows two broad categories. It cannot identify specific conditions or assign clinical categories such as O-RADS.
- It will still return scores for images that are not ovarian ultrasounds.
Intended use
Research and education only. Cathra is not for screening, diagnosis, triage, or any medical decision, and it is not intended for patients. Ultrasound images must be reviewed by a qualified medical professional.
Ownership
Cathra’s model, code, and design are proprietary and protected by copyright. You may not copy, download, redistribute, modify, or reverse-engineer them without written permission. Requests: paultaaron@icloud.com.
Medical & legal disclaimer
Cathra is a research and educational demonstration only. It is not a medical device, is not FDA-cleared or approved, and does not diagnose, treat, cure, prevent, detect, confirm, or rule out ovarian cancer or any other medical condition. Using it does not create a doctor-patient relationship. Results are computer-generated similarity scores provided “as is,” without warranties of any kind, and may be inaccurate or incomplete. Do not use them to make, delay, or disregard any medical decision, and always consult a qualified, licensed medical professional. You use Cathra entirely at your own risk.
Privacy
The image is processed on your device and is never uploaded or stored. See the privacy policy for details.