Improving Multi-task Learning For Breast Cancer Detection
Trung Chi Nguyen, Thong Dinh Nguyen, Thi Thanh Sang Nguyen · 2024
Breast cancer is a significant health issue affecting women worldwide, and mammography is the most effective method for screening cancer. In this paper, we propose a novel approach that uses a deep learning model trained on a large dataset of mammograms, and by using a multi-task learning approach, the model can learn to extract features simultaneously relevant to both classification tasks of (1) Breast Imaging Reporting and Data System (BI-RADS) and (2) density. Besides, we also apply augmentation techniques, e.g., Mixup and concatenating, to enrich the dataset and improve the performance of the model. The experimental results have shown that our proposal is promising, compared with the traditional ResNet variants. Our work contributes to the development of advanced machine-learning methods for breast cancer detection and provides a possible avenue for improving patient outcomes.