AI shows promise for radiotherapy-associated cardiovascular toxicity (CVT) prediction and imaging, a new systematic review has found.
Researchers warned, however, that it is too underdeveloped and lacking in robust validation in the relevant field for routine clinical implementation.
CVT Post-Radiotherapy
Radiotherapy increases long-term CVT in cancer survivors, remaining a major concern through its contribution to morbidity and mortality.
AI, researchers reported, could improve risk prediction and clinical planning of radiotherapy, however its role has thus far been unclear.
AI Performance
Authors included 65 studies in the review, that analysed AI prediction models and cardiovascular imaging applications.
Deep learning was the most common method, with 69% of studies implementing, and exhibited the greatest predictive performance of CVT.
Calibration assessment, at 10% inclusion, and external validation, at 19%, were limited.
The meta-analysis gave an overall predictive model accuracy of 0.83.
Imaging models performed highly for larger cardiac structures, whilst, in contrast, coronary artery segmentation posed a challenge.
Caveats to AI Implementation
Notably, 97% of predictive and 82% of imaging studies were rated at high-risk of bias.
Reports of high AI performance were also limited by heterogeneity and incomplete reporting.
Further, methodological quality was limited.
Researchers called for standardised endpoints in this field of research, alongside prospective clinical integration.
Future of AI in Radiotherapy
Nonetheless, AI applications in radiotherapy are rapidly expanding and show a promising ability to improve cardiac risk prediction and imaging-based radiation dose optimisation.
In dose optimisation, particularly for coronary artery structures, shared and high-quality annotated imaging datasets must be developed.
Collaboration between radiation oncologists, cardiologists, radiologists, imaging scientists, and data scientists is paramount to the successful implementation of AI tools in this sphere, researchers reported.
Implementation must also be centred around clinically relevant endpoints and survivorship care models.
Reference
Salama V et al. A systematic review of artificial intelligence in radiotherapy associated cardiovascular toxicity. Cardiooncology. 2026;DOI:10.1186/s40959-026-00528-5.
Featured image: Mark Kostich on Adobe Stock
- Author :
