Deep learning-derived 12-lead electrocardiogram-based genotype prediction for hypertrophic cardiomyopathy: a pilot...

Area of Science:

• Cardiology
• Medical Artificial Intelligence
• Genetics

Background:

• Hypertrophic cardiomyopathy (HCM) presents significant psychosocial and ethical challenges for patients.
• Establishing genetic probability before genetic testing could benefit HCM patients.
• Current predictive methods for HCM genotypes have limitations.

Purpose of the Study:

• To develop a simple tool for genotype prediction in hypertrophic cardiomyopathy (HCM) patients.
• To evaluate the performance of a convolutional neural network (CNN) against conventional scoring systems.
• To explore the utility of the CNN in sudden cardiac death (SCD) risk stratification for HCM.

Main Methods:

• A CNN was developed using 12-lead electrocardiogram (ECG) data from 124 HCM patients who underwent genetic testing (GT).
• The CNN model was externally validated on a separate HCM cohort (n=54) and compared with Mayo and Toronto scores.
• A third HCM cohort (n=76) was used to assess SCD risk (HCM risk-SCD) based on predicted genotypes, with Score-CAM for visualization.

Main Results:

• Overall, 45% (80/178) of HCM patients were genotype-positive.
• The CNN achieved an AUC of 0.89 on the test set, significantly outperforming the Mayo (0.69) and Toronto (0.69) scores (p < 0.001).
• Patients predicted as genotype-positive exhibited significantly higher HCM risk-SCD scores compared to predicted genotype-negative patients (p < 0.01).

Conclusions:

• The developed CNN shows a promising capability for accurate genotype prediction in HCM patients using ECG data.
• The model effectively aids in risk stratification for sudden cardiac death in hypertrophic cardiomyopathy.
• Limb leads were identified as key contributors to the CNN's prediction accuracy via visualization.