A systematic review and meta-analysis on the performance of convolutional neural networks ECGs in the diagnosis of hypertrophic cardiomyopathy

Abstract

Introduction

Hypertrophic cardiomyopathy (HCM) is a leading cause of sudden cardiac death in younger individuals. Accurate diagnosis is crucial for management and improving patient outcomes. The application of convolutional Neural Networks (CNN), a type of AI modeling, to electrocardiogram (ECG) analysis, presents a promising and optimistic avenue for the detection of HCM. We conducted a meta-analysis to assess the effectiveness of CNN models in diagnosing HCM through ECG.

Methods

MEDLINE, Embase, and Cochrane were searched up to August 12, 2024, focusing on CNN ECG-based HCM detection models. The outcomes were sensitivity, specificity, and SROC. Pooled proportions were calculated using a random-effects model with 95 % confidence intervals (CIs), and heterogeneity was assessed using the I² statistics. This study was registered on PROSPERO protocol CRD42024581925.

Results

Our analysis included 16 studies with ECG data from 513,972 patients. The AI algorithms employed CNNs for ECG interpretation. Sixteen studies contributed to the qualitative analysis, while seven studies for the pooled SROC with an 11 % false positive rate, with a sensitivity of 89 % (95 % CI 86–92 %) and a specificity of 88 % (95 % CI 81–93 %).

Conclusion

AI-driven ECG interpretation shows high accuracy and sensitivity in detecting HCM, though the modest PPV suggests that AI should be integrated with clinical evaluation to enhance reliability, particularly in screening settings.