Prognostic significance of artificial intelligence quantified late gadolinium enhancement in hypertrophic cardiomyopathy
Y Jeong, C Lee, E KongAbstract
Background
Late gadolinium enhancement (LGE) on cardiac magnetic resonance (CMR) reflects myocardial fibrosis and serves as a strong prognostic marker in hypertrophic cardiomyopathy (HCM). However, conventional manual LGE quantification is labor-intensive and limited by inter-observer variability, hindering its routine use. Artificial intelligence (AI)-based quantification enables automated and reproducible assessment of LGE, potentially improving clinical applicability for risk stratification.
Purpose
This study aimed to evaluate the prognostic value of AI-based LGE quantification in patients with HCM and to determine an optimal cutoff value for predicting adverse cardiovascular outcomes.
Methods
We retrospectively analyzed 142 patients with HCM (mean age 58.5 ± 13.7 years, 71% men) who underwent CMR between 2015 and 2023. LGE was quantified using the 6-standard deviation method with an AI segmentation platform (Myomics, Phantomics Inc., Seoul, Korea). Patients were categorized into high (≥15%) and low (<15%) LGE groups. The primary endpoint was a composite of cardiovascular death and sudden cardiac death (SCD)–equivalent events. Receiver operating characteristic (ROC) analysis was used to identify an optimal LGE cutoff.
Results
During a median follow-up of 59 months, 12 patients (8.5%) experienced the primary endpoint. Event rates were higher in the high LGE group (21.2% vs. 4.6%; p = 0.0067). In multivariable Cox regression, LGE percentage remained independently associated with the primary outcome (adjusted HR 4.67, 95% CI 1.43–15.30, p = 0.011). ROC analysis identified an optimal cutoff of 8% (AUC 0.821, sensitivity 100%, specificity 59%), and no events occurred below this threshold, suggesting a potential "safety margin." Patients with LGE ≥8% had significantly higher risks of both primary and SCD/SCD-equivalent events (p < 0.001).
Conclusions
AI-quantified LGE burden was an independent predictor of adverse cardiovascular outcomes in HCM. Automated AI-based analysis enables rapid, standardized, and reproducible assessment of myocardial fibrosis and may refine SCD risk stratification. An AI-derived LGE threshold of 8% appears to identify a low-risk subgroup, supporting the integration of quantitative AI analysis into clinical decision-making for HCM management.