Deep learning can predict cardiovascular events from liver imaging.

Background & Aims: Cardiovascular mortality remains the leading cause of death and a significant source of morbidity, with metabolic alterations being key etiological factors. As the main metabolic organ, the liver could predict prodromal changes associated with increased cardiovascular risk. However, quantifying this risk remains challenging. This study explores the use of transformer neural networks on liver magnetic resonance imaging (MRI) data to enhance cardiovascular risk prediction. Methods: Using the extensive collection of liver MRIs in the UK Biobank, we developed a feature extractor with a vision transformer backbone trained in a self-supervised manner. This encoder was then used to predict cardiovascular outcomes from liver MRI scans. Unlike traditional methods, no manual feature selection was required, minimizing bias. Performance was assessed via fivefold cross validation, where predicted risk scores were compared against actual cardiovascular outcomes. Results: The model was trained on 44,672 liver MRIs. In the fivefold cross-validation predicting major adverse cardiac events, the mean AUC was 0.70 with a 95% CI of 0.69-0.72 and Conclusions: Vision transformer-based models demonstrate promise as quantifiable biomarkers for cardiovascular risk assessment by capturing subtle metabolic and vascular information from liver MRI scans. These findings highlight their strong predictive performance and potential value in risk stratification. Further prospective studies and external validation will be required to establish their clinical utility. Impact and implications: Our study demonstrates that deep learning applied to liver MRI can predict cardiovascular risk, highlighting the role of the liver as a metabolic indicator of early cardiovascular disease. These findings are significant for clinicians and researchers seeking non-invasive, imaging-based biomarkers for cardiovascular risk stratification, particularly in patients who might not yet exhibit overt symptoms. If validated in prospective studies, this approach could enhance current risk assessment models, allowing for earlier and more personalized interventions in high-risk individuals. However, further validation is necessary before clinical implementation, ensuring broad applicability and integration into existing prevention frameworks.