In the era of precision medicine, cardiovascular risk prediction is evolving beyond traditional clinical scores and static imaging markers. The recent study published in the European Heart Journal by Miller et al.  “Deep learning-quantified body composition from PET/CT and cardiovascular outcomes: a multicentre study” exemplifies how artificial intelligence can transform standard imaging into a powerful tool for metabolic profiling and prognostication.

The authors present a multicentre analysis involving more than 10,000 patients undergoing PET myocardial perfusion imaging (MPI), in which low-dose CT attenuation correction scans were retrospectively processed using deep learning algorithms to quantify body composition, specifically skeletal muscle, bone, and several adipose tissue compartments (subcutaneous, epicardial, visceral, and intermuscular fat). These volumetric and density-based metrics were then correlated with long-term cardiovascular outcomes, including all-cause mortality and myocardial infarction.

What makes this study particularly impactful is its pragmatic approach. The CT images analyzed were not acquired for body composition assessment, but rather for attenuation correction during PET MPI, a routine component of clinical imaging workflows. Nonetheless, through advanced segmentation and automated analysis, the authors demonstrated that these images harbor a wealth of prognostically relevant information.
The findings are striking: high visceral adipose tissue density and low skeletal muscle volume were independently associated with increased risk of death or myocardial infarction, even after adjustment for age, sex, body mass index , coronary artery calcium, myocardial flow reserve, and other established predictors. The integration of these novel markers significantly improved model discrimination, underscoring their potential clinical utility. Of note, subcutaneous adipose tissue volume showed a paradoxical protective effect, highlighting the heterogeneity of adipose depots and reinforcing the concept that not all fat is equal in terms of cardiometabolic risk.

Beyond these associations, the study provides a robust framework for opportunistic risk stratification. Physicians could, in theory, receive automated body composition reports alongside standard PET MPI results, allowing for a more holistic assessment of patient health, including sarcopenia, myosteatosis, and ectopic fat accumulation, all of which are increasingly recognized as key components of cardiovascular vulnerability.
From a clinical perspective, these findings are particularly relevant in patients with cardiometabolic syndrome, obesity, or frailty, populations where standard imaging may underestimate risk and where tailored interventions (e.g., nutritional support, cardiac rehabilitation, anti-inflammatory strategies) could be beneficial. Moreover, the ability to derive this information from existing data sets, without additional radiation or cost, makes this approach highly scalable and attractive for real-world application.

In summary, this study exemplifies the power of AI to extract hidden layers of value from routine cardiovascular imaging. It encourages cardiologists to look beyond the coronary arteries and consider the broader systemic context captured by each scan. As we move toward more integrated and data-driven care, body composition analysis via deep learning could become a cornerstone of personalized cardiovascular risk assessment.
We highly recommend this article to clinicians and researchers interested in cardiometabolic imaging, AI applications in cardiovascular medicine, and the development of new prognostic tools that bridge metabolic and structural domains.