Computer Science

Abstract: Breast cancer is one of the most common cancers found primarily in women, affecting 1 in 8 women in Canada over their lifetime, according to the Public Health Agency of Canada. The early diagnosis of breast cancer can reduce the spread of cancer cells and the risk of death. Machine learning (ML) systems can help healthcare professionals diagnose breast cancer with better accuracy by seeing the underlying patterns in the images that the radiologist might find difficult to detect early on. While deep learning (DL) models can obtain high accuracies, their black-box nature inhibits their ability to explain how they came to their conclusion and their regions of interest (ROI), creating distrust between DL systems and the radiologist. This research employs two CNN model architectures - AlexNet and Houby & Yassin's CNN – to classify recently obtained mammographic images from the KAU-BCMD dataset into six BI-RAD categories (0 - 6). To resolve the problem, SHAP, an explainability approach based on Shapley values, has been used to assess the role that each feature has towards the prediction process. There were two SHAP variants that were applied: PartitionSHAP and DeepSHAP. The method of employing the dataset through the models is by 1) preprocessing the dataset, 2) classifying the dataset in the BI-RADs 0-6, and 3) obtaining the SHAP plots identifying the model's ROIs. The proposed methods resulted in high-performance metrics for AlexNet and Houby & Yassin's CNN with an AUC and F1-score of 0.983 and 0.888, and 0.982 and 0.891, respectively.