Self-Supervised AI: A New Era in Biomedical Image Analysis

Speaker:

Aristotelis Tsirigos, PhD

Professor, Departments of Medicine and Pathology

Director. Applied Bioinformatics Laboratories

Abstract:

In this talk, Professor Tsirigos will explore innovative advancements in the field of computational pathology, focusing on the application of self-supervised artificial intelligence. His presentation is centered on a recent study published in Nature Communications, where his team introduced a self-supervised learning framework to analyze unannotated pathology slides. The aim of this approach is to address the inherent limitations of traditional supervised AI methods, which rely heavily on expert annotations that are time-consuming, costly, and susceptible to human biases. Dr. Tsirigos’s methodology involves Histomorphological Phenotype Learning (HPL), a pipeline that automatically identifies distinct tissue patterns from whole-slide images (WSIs) without requiring manual labeling. By segmenting these large WSIs into tiles, the AI autonomously organizes them into clusters, representing distinct morphological features. These clusters form an atlas that provides a visual and interpretable representation of various tumor phenotypes, ranging from benign to malignant states. The study demonstrated the model’s capacity to predict crucial clinical outcomes, such as overall survival and recurrence-free survival, with high accuracy. The AI not only identified known cancer phenotypes but also revealed new patterns with prognostic implications. Dr. Tsirigos’s work bridges the gap between histopathology, genomics, and clinical data, enhancing diagnostic accuracy and paving the way for precision medicine. By offering unbiased, scalable, and interpretable insights, this self-supervised AI model marks a significant step towards revolutionizing cancer diagnosis and treatment planning, providing a reliable second opinion for oncologists and ultimately improving patient outcomes in a tangible and promising way.

Dr. Tsirigos received his B.S. from the National Technical University of Athens, Greece, in 1998 and his Ph.D. in Computer Science from Courant Institute of New York University in 2006. Afterward, he spent eight years with IBM's Bioinformatics and Pattern Discovery Group. In 2015, he accepted an offer from NYU’s School of Medicine. As the Director of the Applied Bioinformatics Laboratories, he leads a team of over 20 computational biologists and data scientists. He has co-authored more than 180 peer-reviewed papers concerning high-impact cancer epigenetics studies, chromatin organi-zation, high-throughput single-cell transcript-tomic analysis, and precision diagnostics using deep learning.