Machine Learning-Based Clustering Using a 12-Lead Electrocardiogram in Patients With a Implantable Cardioverter...

Area of Science:

• Cardiology
• Biomedical Engineering
• Data Science

Background:

• Implantable cardioverter-defibrillators (ICDs) are crucial for reducing mortality in high-risk cardiovascular patients with ventricular arrhythmias.
• Machine learning (ML) shows promise in arrhythmia research, but its use in predicting ventricular arrhythmias in ICD patients is unexplored.

Purpose of the Study:

• To predict and stratify ventricular arrhythmias requiring ICD therapy.
• To utilize 12-lead electrocardiograms (ECGs) in patients with an ICD for ML-based risk stratification.

Main Methods:

• Retrospective analysis of 200 adult patients with ICDs.
• Application of unsupervised learning techniques (K-means, hierarchical clustering) and dimensionality reduction on 12-lead ECG data.
• Silhouette coefficient used to determine optimal clustering method and number of clusters.

Main Results:

• Hierarchical clustering into 3 clusters demonstrated the highest accuracy (silhouette coefficient=0.585).
• Kaplan-Meier analysis revealed significant differences between the 3 identified clusters (P=0.026).
• 29.5% of patients received appropriate ICD therapy during a mean follow-up of 2,953 days.

Conclusions:

• ML-based clustering of 12-lead ECGs holds potential for ventricular arrhythmia risk stratification in ICD patients.
• Further multicenter research is needed to refine ICD indications and validate findings.