Contrastive Integrated Gradients: A Feature Attribution-Based Method for Explaining Whole Slide Image Classification

- The introduction of Contrastive Integrated Gradients (CIG) marks a significant advancement in Whole Slide Image analysis for computational pathology, enhancing interpretability and model trustworthiness. CIG effectively highlights class-discriminative regions, addressing the limitations of existing methods like Integrated Gradients in high-resolution contexts. This development is crucial for improving AI-assisted diagnostics, particularly in distinguishing tumor subtypes, which is essential for effective treatment planning.