Emergence of digital pathology has led to a leap in availability of digitalized whole slide images, providing a wealth of data for developing computational methods for interpreting the images. Realizing the full potential of artificial intelligence based computational pathology requires high-performance computing resources. Here, we study the use of generative and predictive modeling using high-performance computing and modern deep learning based artificial intelligence for histopathology. We develop foundational histology models using self-supervised learning for massive public domain datasets. Further, we extend the possibilities for using unstained, label-free tissue images, reducing the hazardous chemical burden for environment, and enabling tissue interpretation beyond the capabilities of human vision. Further, we will extend cross-modality transforms from label-free histology towards new applications in histogenomic and -proteomic analysis in cancer.

2024

2026