Mammography Quality Evaluation and Model Interpretation Based on CNN-based Inframammary Fold Classification

The Inframammary fold is the lower breast boundary connected to the underlying chest wall. The presence of an inframammary fold is one of the positioning criteria for mammography quality evaluation. As the inframammary fold appeared on a MedioLateral Oblique- view mammogram means that posterior breast tissue is well-presented. This paper introduces an inframammary fold classification method based on a convolutional neural network (CNN) for mammography quality evaluation. The classifier utilizes three CNN models separately, including, AlexNet, GoogLeNet, and Vgg-16. In addition, we implement the two-category and the three-category classifications. Our method extracts a class activation map and then analyzes it to interpret what features the classifier observes. The experiment results demonstrate that the proposed method outperforms the previous work. The CAM depicts why the classifier efficiently determines the presence of an inframammary fold.