Project 1:Al-driven patho-genomic predictive models for personalized treatment of cervical and anal cancers.

University Of Miami School Of Medicine, Coral Gables FL

Investigators

Abstract

Project 1 within our SPORE focuses on the development of AI-driven computational pathology models to address population differences in cervical cancer (CC) and anal cancer (AC). Our novel approach involves the development and validation of population specific and tailored prognostic and treatment response prediction models across large, unique, multi-institutional and clinical trial datasets. By integrating explainable artificial intelligence (XAI) and computer vision tools, we aim to identify population-specific tumor microenvironment morphometric features, including immune cell density, architecture, and collagen arrangement, to stratify patients and predict treatment response in a population specific manner, using just routine H&E digitized slide images. Our innovative approach, grounded in the analysis of standard pathology slide images, allows for efficient, cost-effective disease and patient risk stratification without requiring expensive molecular testing. By relying solely on the use of digital pathology images, the approach is both readily translatable and scalable. The method utilizes digital pathology images corresponding to biopsy and surgically excised specimens of CC and AC, along with corresponding global ancestry, and other demographic related information. Using novel and interpretable computer vision tools, we will first identify potential morphologic and cellular differences in the disease phenotype across different populations (Hispanic, European American and African American patients) and then leverage these differences to create population tailored models for predicting disease outcomes and treatment response across different modalities including chemotherapy, radiation and immunotherapy. Additionally, the project includes auditing algorithmic bias and validating population-specific models on large datasets. This effort will generate robust, interpretable, prognostic and predictive models that facilitate personalized treatment strategies, providing AI-driven solutions that are transparent and clinically translatable to different population groups.