PCOS XAI Ultrasound Dataset
An ovarian ultrasound set simulating real clinical data, ideal for learning how to prepare complex medical data.
11,784 raw ultrasound JPEG images, with artifacts and various resolutions
CC0: Public Domain
Description
The dataset PCOS XAI Ultrasound contains 11,784 ovarian ultrasound images, annotated manually, divided into two categories: cases of PCOS (polycystic ovary syndrome) and healthy ovaries. The data is purposely raw to simulate the challenges of real clinical data, such as duplicates, resolution variations, class imbalances, and visual artifacts.
What is this dataset for?
- Training XAI models to detect PCOS
- Testing medical data cleaning pipelines
- Evaluate duplicate detection and standardization methods
Can it be enriched or improved?
Yes, this dataset can be enriched with finer annotation (disease stage, image quality), or transformed for multi-class projects. It can also be used as a basis for training cleaning models or for comparative analysis between annotated and raw versions.
🔎 In summary
🧠 Recommended for
- Medical data science students
- Diagnostic imaging researchers
- Explainable AI engineers
🔧 Compatible tools
- PyTorch
- TensorFlow
- OpenCV
- FastAI
- Roboflow
💡 Tip
Leverage duplicates and errors to create a robust automated cleaning pipeline.
Frequently Asked Questions
Is this dataset ready to use for a classification model?
No, it is deliberately disordered to train users to clean and structure data before modeling.
Can this dataset be used for scientific publications?
Yes, the CC0 license allows it, but it is recommended to mention the limitations of the dataset and the cleaning efforts made.
What is the best approach to dealing with class imbalance?
Techniques such as targeted augmentation, oversampling (SMOTE) or loss weighting are recommended.
