11.00 – 11.30 Scott doyle - Building an AI School for Pathology: Workflow for Human-AI Interfaces
As machine learning and artificial intelligence becomes more widely used in pathology, it is crucial to develop toolsets and teaching opportunities for future pathologists to understand and interact with these systems. At the University at Buffalo, we have launched a new graduate program that seeks to integrate computational, quantitative tools – including AI – into pathology and microscopy training. This presentation will discuss our ongoing efforts to engage pathology students and residents, allowing them to interact with AI systems for object detection and segmentation on whole slide imaging. We will talk about our work in developing active learning approaches for rapid and efficient annotation, which includes presenting the results of deep learning segmentation to the annotators for correction. This results in a joint learning experience: as the AI receives quality data, the human student can see how the AI “thinks” about the data. We believe this will lead to wider acceptance, understanding, and competent adoption of these methods by clinical pathologists.
CV Dr. Scott Doyle received his PhD in Biomedical Engineering from Rutgers University in 2011, where he worked on developing computational pathology tools for quantifying structure of prostate and breast cancers in clinical samples of H&E-stained tissue. He worked for a start-up company to develop these methods in the commercial space before coming to the University at Buffalo in 2014 as an Assistant Professor. His lab focuses on developing computational approaches for understanding biology and disease, with the hypothesis that structural data (as captured by biomedical imaging) is minable, quantifiable data that can be used in a statistical framework. Recently he has been developing methods to understand how modern deep learning and AI methods “learn” about the image space they are trained on, and is working on streamlining an efficient process for training and obtaining annotations.