Multi-View Attention Network to Improve Breast Cancer Detection

Published

Medical Imaging with Deep Learning (2024)

Authors

Wen Tai¹, PinJui Huang¹, and Dongmyung Shin²

Affiliations

¹Marketech International Corp, Taipei, Taiwan
²Artificial Intelligence Engineering Division, Radisen Co. Ltd

Abstract

Breast cancer is the most prevalent cancer in women, and mammography is an effective imaging modality for detecting it in its early stages. However, identifying tumors in mammograms is challenging, and many AI algorithms have been proposed to assist radiologists in detecting them. This study focuses on demonstrating the potential of a multi-view attention network for breast cancer detection by investigating the change in the detection performance depending on the types of attention (no, single-view, or multi-view attention), image resolution (low or high), and backbone network (ResNet50 or HRNet). The experiment results showed that the detection performance of a high-resolution, multi-view attention network with an HRNet backbone was better than the other networks with different configurations, suggesting that multi-view attention has benefits in detecting masses on mammograms.