Pathology is a digital technology for the capture and analysis of high-resolution Whole Slide Images (WSIs). WSI entails scanning glass pathology slides, typically containing tissue samples, into a digital format. This facilitates comprehensive digital viewing, analysis, and interpretation by pathologists. Despite the undeniable utility of this method, the inherent time-consuming and labor-intensive nature of traditional pathology image analysis, which is heavily reliant on expert pathologists, is well-known. Recent years have seen a surge in research aiming to address these challenges through the development of automated systems using machine learning approaches. While promising, such systems often exhibit bias towards specific datasets and struggle to transition effectively to real-world scenarios. For this reason, there is a need to establish a uniform machine learning training approach for generating more robust and consistent results. In this paper, I introduce a contrastive learning-based multi-task whole slide image classification system. The proposed system excels at extracting consistently reliable features for identical cancer categories, thereby enhancing accuracy in downstream tasks. Through extensive experiments, the results demonstrate that the proposed system outperforms pre-existing state-of-the-art machine learning models. I expect that the proposed system can significantly contribute to pathologists by offering valuable cancer screening capabilities in WSIs.
Sean Hwang (Fri,) studied this question.