Deep Learning Based Binary Classification of Invasive Ductal Carcinoma: A Comparative Study on CNN and VIT Models

This research concerns binary classification of Invasive Ductal Carcinoma (IDC) using two deep learning models, Convolutional Neural Networks (CNNs) along with Vision Transformers (ViTs). We use three different datasets, which consist of both histopathological and ultrasound images for training and comparison among the models. CNNs outperformed ViTs in terms of state-of-the-art accuracy and computational efficiency with a maximum accuracy up to 97.03% for certain datasets, while ViTs performed extremely well with an impressive recall value but could not perform as well as CNNs leading towards more false positives. The performances of CNNs are reported to be more accurate in terms of both resource-constrained and IDC classification. We aim to further refine the latter by integrating CNNs and ViTs within hybrid models for increased performance, in addition to explore the explainable AI to increase clinical trustworthiness and adoption.