Histopathology image classification using supervised contrastive deep learning

Cancer remains a significant global health problem, with the cornerstone of its diagnosis being histopathological image analysis, which can be prone to human subjectivity and error. This paper introduces CHistNet, a novel framework tailored for histopathology image classification. CHistNet leverages the advantages of contrastive learning strategies, especially beneficial for imbalanced data scenarios, in combination with a cross-entropy loss function. Our approach focuses on a sequential methodology, transitioning from feature to classifier learning, truly reflecting the core principles of contrastive learning to produce superior feature representations. We extensively tested our method on five publicly available datasets from various histopathology domains, achieving state-of-the-art accuracy. This research provides a valuable tool for enhancing cancer diagnosis through increased precision. The source code of this study is available at: https://github.com/Mamunur-20/CHistNet.