Automatic Detection of Left Ventricular Dilatation and Hypertrophy from Electrocardiograms Using Deep Learning

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Left ventricular dilatation (LVD) and left ventricular hypertrophy (LVH) are key risk factors for heart failure.
• Early detection of LVD and LVH can significantly improve heart failure screening and patient outcomes.

Purpose of the Study:

• To investigate the efficacy of deep learning models in detecting LVD and LVH using 12-lead electrocardiograms (ECGs).
• To compare the performance of deep learning models against traditional machine learning methods and conventional ECG criteria.

Main Methods:

• Development of deep learning and machine learning models using ECG and echocardiographic data from 18,954 patients.
• Evaluation of model performance using area under the receiver operating characteristic (AUROC) curves, sensitivity, specificity, and accuracy.
• Comparison of deep learning model performance against logistic regression, random forest, and conventional ECG criteria for LVH.

Main Results:

• Deep learning models achieved significantly higher AUROCs for detecting both LVD (0.810) and LVH (0.784) compared to machine learning models and conventional ECG criteria.
• The deep learning model for LVD detection showed a significantly higher AUROC than logistic regression (0.770) and random forest (0.757).
• For LVH detection, the deep learning model's AUROC (0.784) was significantly superior to logistic regression (0.758), random forest (0.716), and conventional ECG criteria.

Conclusions:

• Deep learning demonstrates a powerful capability for detecting LVD and LVH directly from standard 12-lead ECGs.
• This approach offers a promising, non-invasive tool for enhancing the early screening and diagnosis of heart failure risk.
• The findings suggest integrating deep learning into ECG analysis could revolutionize cardiovascular risk assessment.