In a recent episode of OncChats, titled “The Future of Pathology with AI,” leading experts discussed the transformative role of artificial intelligence (AI) in modern pathology. Dr. Toufic Kachaamy, chief of medicine at City of Hope, Dr. Madappa Kundranda, chief of cancer medicine at Banner MD Anderson Cancer Center, and Dr. Kun-Hsing Yu, an assistant professor at Harvard Medical School, provided insights into how AI is reshaping cancer diagnosis.
Dr. Kachaamy opened the discussion by welcoming the panel and emphasizing the importance of the topic. He introduced Dr. Yu, who has pioneered the first fully automated AI algorithm for extracting extensive features from whole slide histopathology images. Dr. Yu”s work has led to the discovery of molecular mechanisms that explain the microscopic characteristics of tumor cells, as well as the identification of new cellular morphologies linked to patient outcomes. His lab focuses on integrating multiomics data, including genomics and proteomics, with quantitative histopathology to anticipate clinical phenotypes.
Dr. Yu elaborated on the current state of pathology and the integration of AI. He highlighted that his research group is dedicated to developing innovative AI methodologies to analyze pathology data, aiming to enhance clinical practices. They have created foundation models using a vast array of data to train AI systems in recognizing essential pathology patterns pertinent to diagnosis and prognosis. Dr. Yu expressed optimism about the successful application of these platforms in various clinical scenarios, including ongoing prospective studies.
Dr. Kachaamy then posed a critical question: Can AI currently diagnose cancers using standard pathology slides? Dr. Yu responded affirmatively, noting that researchers have been working for decades to create quantitative methods for analyzing whole slide digital pathology images. He mentioned that their recent foundation model can accurately diagnose over 19 different cancer types, distinguishing cancerous regions from adjacent benign tissues and classifying various cancer types and subtypes. He asserted that the diagnostic accuracy achieved by their AI system is comparable to that of expert pathologists, underscoring the potential of AI when trained with appropriate data and architecture.
The discussion highlighted the promising future of pathology with AI, emphasizing the potential for improved diagnostic accuracy and patient outcomes. As AI continues to evolve, its integration into pathology could lead to significant advancements in cancer care.
