Improving localization-based approaches for breast cancer screening exam classification
This work addresses breast cancer screening for medical diagnosis, providing an incremental improvement with interpretable predictions.
The paper tackled breast cancer screening exam classification by training a localization-based deep CNN on over 200,000 exams, achieving an AUC of 0.919 for malignancy prediction and reducing the baseline error rate by 23%.
We trained and evaluated a localization-based deep CNN for breast cancer screening exam classification on over 200,000 exams (over 1,000,000 images). Our model achieves an AUC of 0.919 in predicting malignancy in patients undergoing breast cancer screening, reducing the error rate of the baseline (Wu et al., 2019a) by 23%. In addition, the models generates bounding boxes for benign and malignant findings, providing interpretable predictions.