University of Pennsylvania Researchers Develop AI Tool for Enhanced Cancer Diagnosis and Treatment Decision-Making
Researchers at the Perelman School of Medicine at the University of Pennsylvania have unveiled an innovative AI tool named iStar, designed to revolutionize cancer diagnosis and treatment decisions. iStar utilizes artificial intelligence (AI) machine learning to automatically identify tumors and various challenging cancer types, while also predicting candidates for immunotherapy. The tool, functioning akin to a human pathologist, analyzes tissue biopsies with near single-cell resolution, providing crucial insights for precision oncology.
Functionality of iStar
iStar generates spatial transcriptomics (ST) data, offering positional information and the first step in gene expression analysis for intact cells or tissues. It combines histology, the study of microscopic anatomy, with a self-supervised learning (SSL) deep learning algorithm. The AI model, a hierarchical vision transformer (HViT), is pre-trained on unlabeled data to generate data labels. iStar’s histology feature extractor then extracts features from images, predicting super-resolution gene expression based on these features. The resulting gene expression information enables tissue segmentation, contributing to enhanced cancer detection.
Detection Across Multiple Cancer Types
The research team assessed iStar using datasets for various cancer types, including breast (including HER2-positive), prostate, colorectal, and kidney cancers. The AI system demonstrated accurate predictions of super-resolution gene expressions across diverse datasets, showcasing its potential for broad applications in cancer diagnosis.
Immunotherapy Biomarker Prediction
iStar’s capabilities extend beyond cancer detection to include the successful identification of immune cell clusters called tertiary lymphoid structures (TLS). TLS presence has been associated with favorable responses to immunotherapy in solid tumors. By predicting these potential biomarkers, iStar contributes valuable insights for clinicians considering immunotherapy as part of the treatment strategy.
The integration of artificial intelligence machine learning, genomics, imaging, and biology in iStar provides clinicians with timely and actionable insights. This interdisciplinary approach enhances the understanding of cancer at a molecular level, supporting precision oncology practices and aiming for positive patient outcomes.
The development of iStar represents a significant advancement in the intersection of AI and cancer research. By automating complex tasks traditionally performed by pathologists, iStar not only streamlines the diagnostic process but also contributes to the identification of potential candidates for immunotherapy. The successful application of iStar across various cancer types underscores its potential as a transformative tool in the pursuit of precision oncology and improved patient outcomes.