Deep learning assessment of left ventricular hypertrophy based on electrocardiogram

Area of Science:

• Cardiology
• Artificial Intelligence in Medicine
• Medical Imaging and Diagnostics

Background:

• Current electrocardiogram (ECG) criteria for diagnosing left ventricular hypertrophy (LVH) exhibit limited sensitivity.
• Deep learning (DL) excels at automatic feature extraction from ECGs for cardiac disease detection, yet its application in LVH diagnosis is underexplored.
• There is a need for improved diagnostic tools for LVH detection using standard 12-lead ECGs.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) model for the rapid and effective detection of left ventricular hypertrophy (LVH) using 12-lead electrocardiograms (ECGs).
• To compare the performance of the DL model against established ECG criteria for LVH diagnosis.

Main Methods:

• A deep learning model integrating convolutional neural network and long short-term memory (CNN-LSTM) was developed to detect LVH from 12-lead ECG data.
• Data from 1,863 patients, including echocardiograms and ECGs, were analyzed and divided into training (n=1,120), validation (n=371), and test sets (n=372).
• An additional internal test set of 453 patients was used for performance validation, with models developed for subgroups based on gender and relative wall thickness (RWT).

Main Results:

• The CNN-LSTM model achieved an area under the curve (AUC) of 0.62 (sensitivity 68%, specificity 57%) in the primary test set, outperforming Cornell (AUC: 0.57) and Sokolow-Lyon (AUC: 0.51) criteria.
• Consistent performance was observed in the internal test set 2, with an AUC of 0.59 (sensitivity 65%, specificity 57%).
• The model demonstrated higher efficacy in male patients (AUC: 0.66, sensitivity 72%, specificity 60%) compared to female patients (AUC: 0.59, sensitivity 50%, specificity 71%).

Conclusions:

• A CNN-LSTM model was successfully established for diagnosing LVH using 12-lead ECG, demonstrating superior sensitivity compared to existing criteria.
• This DL-based approach shows potential as a simple, effective screening tool for identifying left ventricular hypertrophy.