We provide code, model weights (EfficientNetB0, MobileNet, Xception, ResNet50 and DenseNet121, for 224*224, 320*320, 448*448 input resolution) and a test set with 300 breast ultrasound images for physician-ai comparasion in this respository.
All models output predictions have been saved in 'model and physicians performance on AI-Physician Comparasion set' dictionary.
3oo test images for physician-ai comparasion are saved in 'Images of AI-Physician Comparasion'.
Pathological results and birads classes of these 300 images could be found in AI-Physician Comparasion Dataset.xlsx.
Staticial figures are saved in 'Figures'.
15 model weights are available from https://drive.google.com/drive/folders/1Vk7vMobeO0ZqbZwR1ErmDxP4PnAPEzyq?usp=sharing.
To test model performance on 300 image test set, run Model_test.py.
To draw models comparasion figure, run Model Perfomance Comparasion.py.
To draw model-physician performance comparasion of Wilcoxon test figures, run Model-Physician Performance Comparasion.py.
To draw AUROC curve of AI and physicians, run AI-Physician-Comparasion.py