AIDose · The first AI-guided toxicity atlas for safer and more effective abdominal radiation therapy

Radiotherapy (RT) is a critical component of cancer treatment, potentially benefiting approximately 50% of patients. However, less than 80% of patients who could benefit from RT actually receive it in Europe. The risk of toxicities, i.e., the radiation-induced damage of healthy tissues, is one of the main reasons why oncologists may exclude a potentially beneficial RT from the treatment regimen. For example, a recent analysis of the RTOG 0617 trial highlighted that the tumor control benefits of high-dose lung RT are accompanied by higher mortality rates due to extensive heart irradiation. The AIDose project will leverage my work on hepatobiliary toxicity prediction, artificial intelligence (AI) advancements, and medical image analysis to develop the first toxicity risk atlas for thoracic and abdominal organs-at-risk (OARs), i.e., organs located in close proximity to tumors. The atlas is envisioned as a three-dimensional map of OARs with pinpointed anatomical subregions associated with high toxicity risks. The AIDose project has four aims: a) developing AI-driven solutions for morphological and radiomic profiling of OARs; b) extracting and analyzing non-dosimetric clinical features associated with toxicities; c) utilizing AI to identify consistent patterns in radiation doses delivered to OAR subregions, and correlating them with toxicity risks; d) validating the resulting toxicity atlas against multi-hospital RT data. The feasibility of AIDose is supported by my previous research on liver and head-and-neck RT planning, which resulted in publications in key journals in the field and an award from the American Association of Medical Physicists. The use of AI for radiotherapy planning and outcome prediction is believed to be the main area of impact by the American and European Societies for Radiation Oncology. This proposal represents a unique opportunity to advance this highly innovative and clinically important research direction.