Abstract

Computational pathology has emerged as a transformative field, leveraging artificial intelligence to analyze Whole Slide Images (WSIs) and enhance diagnostic accuracy and efficiency. However, traditional approaches often focus solely on learnable features from WSIs, neglecting the critical role of clinical expertise and domain knowledge in interpreting histopathologic entities. In this talk, I will share our recent works in addressing these limitations through the integration of multimodal AI and knowledge-guided frameworks. I will highlight strategies for aligning human-like reasoning with computational methods to enhance diagnostic precision and interpretability. Additionally, I will explore how expert knowledge can be utilized to dynamically adapt pathology foundation models to specific tasks, improving feature representation and performance. These advancements illustrate the transformative potential of multimodal AI to develop precise, interpretable, and clinically impactful diagnostic tools, setting the stage for the next era of computational pathology.