Deep learning of echocardiography distinguishes between presence and absence of late gadolinium enhancement on cardiac magnetic resonance in patients with hypertrophic cardiomyopathy

Comprehensive Summary

Akita et al. developed a deep learning model using apical 5-chamber echocardiographic images to identify the presence of late gadolinium enhancement (LGE) on cardiac magnetic resonance (CMR) in patients with hypertrophic cardiomyopathy. The study included 323 CMR studies from a single tertiary HCM center, of which 160 (50%) had positive LGE and 163 had negative LGE; 273 samples were used for training and 50 for independent testing. The authors also built a reference model using 7 clinical variables, including family history of HCM, maximum left ventricular wall thickness, left ventricular dimensions, left atrial diameter, left ventricular ejection fraction <50%, and resting left ventricular outflow tract gradient. In the test set, the deep learning probability alone had an AUC of 0.74 (95% CI 0.60-0.88), while the combined model significantly outperformed the reference model, with an AUC of 0.86 (95% CI 0.76-0.96) versus 0.72 (95% CI 0.57-0.86; P=0.04). The combined model had a sensitivity of 0.84, specificity of 0.76, positive predictive value of 0.78, and negative predictive value of 0.83. Overall, the authors concluded that adding deep learning analysis of echocardiographic images to clinical variables improved discrimination of LGE on CMR and could help inform decisions about whether to perform gadolinium-enhanced CMR in HCM.

Outcomes and Implications

This model is best understood as a triage aid rather than a replacement for CMR. Its value is that it improved prediction of LGE beyond standard clinical variables, which could help identify HCM patients who are more likely to benefit from gadolinium-enhanced CMR when access, cost, or contraindications make routine scanning harder. However, this was a single-center study with a small test set, and the model predicted only the presence of LGE rather than fibrosis burden or clinical outcomes. Clinically, that supports a workflow role in selecting patients for further imaging, not a stand-alone role in risk stratification or ICD decision-making.