2465 - Artificial Intelligence in Radiation-Associated Cardiovascular Adverse Events: A Systematic Review of Predictive and Imaging Applications
Presenter(s)
V. Salama1, B. M. Godinich2, N. Dunham3, T. Nguyen4, P. M. Lilly1, J. A. Schmidlen5, J. Ryckman1, R. A. C. Siochi6, C. Bianco7, R. R. Raylman4, M. F. Hanna4, and P. Pifer1; 1Department of Radiation Oncology, West Virginia University School of Medicine, WVU Cancer Institute, Morgantown, WV, 2Texas Tech Health Science Center El paso, El Paso, TX, 3West Virginia University, School of Medicine, Charleston, WV, 4Department of Radiology, West Virginia University, School of Medicine, Morgantown, WV, 5West Virginia University School of Medicine, Morgantown, WV, 6Department of Radiation Oncology, West Virginia University Cancer Institute, Morgantown, WV, 7Department of Cardiology and Cardiac Surgery, West Virginia University, School of Medicine, Morgantown, WV
Purpose/Objective(s):
Cardiovascular adverse events (CVAEs) are a major late effect of cancer treatment, including radiation therapy (RT), and an increasing contributor to morbidity and mortality among cancer survivors. Artificial intelligence (AI) has potential to improve early detection, risk stratification, and RT planning to reduce cardiac exposure, but the quality and readiness of the evidence remain unclear. This objective of this study was to systematically evaluate the applications, performance and methodological quality of AI applications in treatment-associated CVAEs among patients undergoing treatment selectively, including RT. We hypothesized that AI models would demonstrate promising predictive and imaging performance, but that many studies would have important quality and methodological limitations.Materials/Methods:
PRISMA-guided systematic review of PubMed, Ovid EMBASE, Cochrane Library, and Web of Science was conducted through October 1, 2025. Eligible studies were original human research in English applying AI to cardiovascular outcomes or imaging in cancer populations receiving RT. Predictive-model studies were assessed using TRIPOD+AI for quality and PROBAST for risk-of-bias. Imaging-AI studies were assessed using CLAIM and QUADAS-2 for quality and risk-of-bias.Results:
Sixty-five studies were included and clustered into two groups: (1) AI prediction of RT-associated CVT (n=31, 48%) and (2) AI-based cardiovascular imaging (n=34, 52%). Deep learning was the most frequent approach (45/65, 69%) especially in imaging and demonstrated highest performance (median AUC=0.82 & sensitivity=0.83) in prediction. Predictive models lacked calibration assessment (3/31, 10%), and external validation (6/31, 19%). TRIPOD+AI adherence averaged 79% (SD 22.68%), while PROBAST rated 97% at high overall risk-of-bias. Imaging models demonstrated strong performance with a median Dice similarity coefficient (DSC) of 0.85 (range 0.76–0.94), especially for larger cardiac structures. Coronary artery segmentation remained challenging. Mean CLAIM adherence was 71%, and 82% of studies were rated high risk of bias by QUADAS-2.Conclusion:
AI applications in radiation-associated cardiovascular adverse events show strong technical performance but remain methodologically limited. Most studies lack external validation, calibration, and low risk-of-bias design. Future research should prioritize standardized CVAE endpoints, rigorous validation, assessment of clinical utility, shared high-quality imaging annotations, and prospective integration into clinical trials to enable safe clinical implementation.