Vision Transformer-Based Breast Mass Diagnosis in Mammography Using Bilateral Information
Tianyu Zeng, Zhang Zhang, Yuwen Zeng, Xiaoyong Zhang, Kei Ichiji, Noriyasu Homma · 2024
In response to the high incidence of breast cancer and the need to enhance automatic diagnosis to assist radiologists, there is an urgent need to develop a system that can aid in the accurate diagnosis of breast cancer. Computer-aided Diagnosis systems have been proposed for the task of breast cancer detection; however, they often encounter challenges in effectively utilizing bilateral images for accurate diagnosis of breast masses. To address this limitation, this study introduces a novel method that leverages a vision transformer to fuse information from both breasts to classify cancerous and normal mammograms. Our study contributes by utilizing bilateral information to diagnose the presence or absence of masses, thereby enhancing the sensitivity and accuracy of disease diagnosis, using the rich breast imaging data in the Digital Database for Screening Mammography. Compared with traditional convolutional neural network-based models, our method improves diagnostic performance and shows better robustness in avoiding registration problems. Experimental results show that our proposed method achieves satisfactory accuracy and reliability in breast mass diagnosis. This improvement leads to more valuable results for clinical decision-making and enhances the diagnostic capabilities of computer-aided diagnosis systems.