Clinical bottom line
Radiation-induced cardiotoxicity is an increasingly recognized complication of thoracic radiotherapy, with potential for significant morbidity and mortality. Early detection is critical to enable timely intervention and mitigate long-term cardiac damage. Emerging evidence supports the use of AI-enhanced imaging modalities, particularly advanced cardiac MRI combined with machine learning algorithms, to identify subclinical myocardial injury and functional impairment before conventional clinical signs or standard imaging changes occur. While promising, these approaches remain largely investigational and require further validation in diverse patient populations and clinical settings.
What the evidence shows
Recent studies have explored multiple AI-augmented imaging techniques for early detection of radiation-induced cardiotoxicity. An Dayeong et al. (2024) demonstrated that advanced machine learning applied to cardiac MRI data in preclinical genetic models can elucidate early markers of radiation-induced cardiac injury, including subtle changes in myocardial strain and tissue characterization not readily apparent on standard imaging. This approach leverages multiparametric MRI sequences combined with AI-driven pattern recognition to improve sensitivity and specificity for early myocardial damage.
Complementing this, Wei Xiaoting et al. (2022) reviewed the role of cardiovascular magnetic resonance imaging (CMR) in detecting cardiotoxicity induced by cancer therapies, highlighting its superior tissue characterization capabilities, including myocardial edema, fibrosis, and functional strain abnormalities. AI integration can enhance image analysis throughput and reduce observer variability, potentially enabling earlier and more reproducible detection of cardiotoxic effects.
Additional modalities under investigation include PET imaging with targeted tracers, such as the [64Cu]Cu-NODAGA-E[(cRGDyK)]2 probe studied in a rat model of doxorubicin-induced heart failure (2014), which may be adaptable for radiation cardiotoxicity assessment by detecting molecular changes preceding structural damage. Similarly, hyperpolarized carbon-13 metabolic imaging has been proposed as a novel method to detect metabolic alterations in irradiated myocardium (2019), though clinical translation remains limited.
Traditional echocardiographic strain imaging, including global longitudinal strain (GLS) and strain rate imaging, remains a valuable clinical tool for early cardiotoxicity detection. Mohty et al. (2017) and Michalaki et al. (2021) provide evidence supporting GLS as a mature biomarker for anthracycline-induced cardiotoxicity, which may be extrapolated cautiously to radiation-induced injury. However, AI-enhanced MRI offers superior spatial resolution and tissue characterization, potentially surpassing echocardiography in early detection.
Biomarker studies summarized by Bulten Bernard Frederik et al. (year not specified) emphasize the complementary role of circulating biomarkers alongside imaging for comprehensive cardiotoxicity surveillance, though imaging remains essential for localizing and quantifying myocardial injury.
Caveats and uncertainty
Despite encouraging preclinical and early clinical data, AI-enhanced imaging for radiation-induced cardiotoxicity is not yet standard practice. Most studies are limited by small sample sizes, animal models, or single-center designs, restricting generalizability. The heterogeneity of radiation protocols, patient comorbidities, and concurrent cardiotoxic therapies complicates interpretation and validation of imaging biomarkers.
AI algorithms require rigorous external validation to confirm reproducibility across different scanners, institutions, and patient populations. Moreover, the incremental clinical benefit of AI-enhanced imaging over established echocardiographic strain methods or conventional CMR parameters remains to be definitively demonstrated.
Cost, accessibility, and the need for specialized expertise in AI and advanced imaging analysis may limit widespread adoption. Integration into clinical workflows and demonstration of impact on patient outcomes through prospective trials are essential next steps.
How this may change practice
If validated, AI-enhanced cardiac MRI and related imaging modalities could transform cardiotoxicity surveillance by enabling earlier, more accurate detection of radiation-induced myocardial injury. This would facilitate timely cardioprotective interventions, personalized radiation planning, and improved long-term cardiac outcomes.
Radiologists and cardio-oncologists should remain informed about evolving imaging techniques and AI applications, considering participation in clinical trials and multidisciplinary collaborations. In the near term, incorporating strain imaging and CMR where available remains prudent, with AI tools as adjuncts pending further evidence.
Ultimately, adoption of AI-enhanced imaging for radiation cardiotoxicity will depend on demonstration of improved diagnostic accuracy, cost-effectiveness, and impact on clinical decision-making and patient prognosis.