Multi-Institution Validation on a Pretrained Vision-Language Prediction Model for Thoracic Radiation-Induced Toxicity
Lung cancer is one of the most common cancers in the U.S., and radiation therapy is a key treatment for many patients. While radiation can be highly effective, some patients experience side effects that significantly impact their quality of life. Identifying those at greatest risk for side effects before treatment begins could help personalize care and improve outcomes. Existing prediction methods are limited and use a fraction of the real-world data available from each patient’s treatment plan. Our research focuses on developing VITAL-RT, a novel artificial intelligence tool designed to predict radiation-related side effects more accurately. By jointly analyzing information collected during routine care, including CT scans, radiation treatment plans, and physicians’ notes, VITAL-RT identifies patterns associated with treatment toxicity to make more accurate predictions than current methods. We will validate VITAL-RT using an independent patient cohort from another institution to test whether it is broadly applicable. Our goal is to provide clinicians with better risk estimates before treatment starts, enabling more personalized treatment planning and closer follow-up when needed without adding major cost or extra tests for patients with lung cancer.
