MamT4: Multi-View Attention Networks for Mammography Cancer Classification

In this study, we introduce a novel method, called MamT$^4$, which is used\nfor simultaneous analysis of four mammography images. A decision is made based\non one image of a breast, with attention also devoted to three additional\nimages: another view of the same breast and two images of the other breast.\nThis approach enables the algorithm to closely replicate the practice of a\nradiologist who reviews the entire set of mammograms for a patient.\nFurthermore, this paper emphasizes the preprocessing of images, specifically\nproposing a cropping model (U-Net based on ResNet-34) to help the method remove\nimage artifacts and focus on the breast region. To the best of our knowledge,\nthis study is the first to achieve a ROC-AUC of 84.0 $\\pm$ 1.7 and an F1 score\nof 56.0 $\\pm$ 1.3 on an independent test dataset of Vietnam digital mammography\n(VinDr-Mammo), which is preprocessed with the cropping model.\n