Deep-Learning-Based Estimation of the Spatial QRS-T Angle from Reduced-Lead ECGs

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• The spatial QRS-T angle is a valuable predictor for sudden cardiac death (SCD) risk stratification.
• Current estimation methods rely on 12-lead electrocardiogram (ECG) systems, which are not suitable for continuous ambulatory monitoring.

Purpose of the Study:

• To develop a novel method for estimating spatial QRS-T angles using a reduced number of ECG leads.
• To facilitate ambulatory monitoring of cardiac health indicators through consumer-grade ECG devices.

Main Methods:

• A deep learning model was designed to identify QRS and T wave vectors from ECG data.
• An innovative loss function was implemented to guide the model in accurately determining vector coordinates in 3D space.
• The model was trained and validated using the extensive PTB-XL dataset, progressively reducing the number of ECG leads.

Main Results:

• Spatial QRS-T angles can be accurately estimated using a subset of four leads: {I, II, aVF, V2}.
• The method achieved acceptable accuracy with absolute mean and median errors of 11.4° and 7.3°, respectively.
• This accuracy is sufficient for identifying abnormal spatial QRS-T angles without compromising patient comfort.

Conclusions:

• The developed deep learning model enables reliable spatial QRS-T angle estimation from reduced-lead ECGs.
• This approach paves the way for ambulatory monitoring of SCD risk using wearable ECG devices.
• Patients at high risk for SCD, including those with chronic cardiac and kidney conditions, stand to benefit significantly.