VISUAL CORRESPONDENCE-BASED EXPLANATIONS IMPROVE CONVOLUTIONAL NEURAL NETWORKSFOR CLASSIFICATIONOF MAMMOGRAMS

Explainable Artificial Intelligence (XAI) classifications are increasingyimportant in convolutional neural networks for the classification of mammograms. In this paper, we propose a novel architecture of Visual Correspondence Based Explanations that improve Convolutional Neural Networks (cnns) for Classification of Mammograms of self-interpretable image classifiers that first explain, and then predict by harnessing the visual correspondences between a query breast cancer X-Ray image and exemplars. In the evaluation of our proposed models, the k-nearest neighbor (knn) classifier improves upon ResNet-18 on Breast Cancer datasets.