Identification of Hypertrophic Cardiomyopathy on Electrocardiographic Images with Deep Learning

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Hypertrophic cardiomyopathy (HCM) affects 1 in 200 individuals and is a primary cause of sudden cardiac death in young adults.
• Current deep learning methods for HCM detection rely on raw electrocardiogram (ECG) voltage data, limiting point-of-care applications.
• Developing accessible methods for HCM screening is crucial for early intervention.

Purpose of the Study:

• To develop and validate a deep learning model for detecting hypertrophic cardiomyopathy (HCM) directly from 12-lead electrocardiogram (ECG) images.
• To overcome the limitations of accessing raw ECG signal data for point-of-care HCM screening.
• To assess the model's performance in diverse clinical settings and external cohorts.

Main Methods:

• A deep learning model was trained on 124,553 ECG images from 66,987 individuals, including patients with confirmed HCM features.
• Model validation was performed internally at YNHH and externally using the UK Biobank cohort.
• Gradient-weighted class activation mapping was employed to localize discriminative ECG signal patterns.

Main Results:

• The model achieved high discrimination for HCM detection, with an area under the receiving operating characteristic curve (AUROC) of 0.96 in internal validation and 0.94 in the UK Biobank cohort.
• A positive screen by the model was strongly associated with CMR-confirmed HCM (OR 102.4).
• Discriminative patterns were localized to the anterior and lateral leads (V4-V5) of the ECG images.

Conclusions:

• A deep learning model effectively identifies hypertrophic cardiomyopathy (HCM) from ECG images, demonstrating excellent discrimination.
• This image-based approach provides an automated, efficient, and accessible strategy for HCM screening.
• The model's external validation confirms its potential for widespread clinical application.