Information bottleneck-based interpretable multitask network for breast cancer classification and segmentation

Junxia Wang, Yuanjie Zheng, Jun Ma, Xinmeng Li, Chongjing Wang, James Gee, Haipeng Wang, Wenhui Huang

Abstract: Highlights•A multi-task learning framework is designed to joint breast cancer classification and tumor segmentation.•An interpretable network based on information bottleneck is proposed for the classification task.•A dual prior knowledge guidance strategy is suggested to improve the task correlation.•Classification and segmentation promote each other during the training process.