QTNet: Deep Learning for Estimating QT Intervals Using a Single Lead ECG.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• QT prolongation is a risk factor for fatal arrhythmias and sudden cardiac death.
• Current QT interval monitoring relies on expert interpretation of 12-lead ECGs, limiting continuous out-of-hospital tracking.
• Advancements in wearable ECG technology and machine learning offer potential for automated QT interval assessment.

Purpose of the Study:

• To develop and validate a deep learning model (QTNet) for accurate QT interval regression from single-lead ECGs.
• To assess the generalizability and robustness of QTNet across diverse datasets.
• To compare QTNet's performance against existing automated ECG analysis methods.

Main Methods:

• A residual neural network, QTNet, was developed for QT interval regression using single-lead (Lead-I) ECG data.
• QTNet was trained in a supervised manner on a large ECG dataset from a U.S. hospital.
• Model performance was evaluated on four independent test sets, including data from different institutions and public datasets.

Main Results:

• QTNet achieved a mean absolute error (MAE) between 9ms and 15.8ms across all test datasets.
• Pearson correlation coefficients for QTNet estimations ranged from 0.899 to 0.914.
• QTNet significantly outperformed a standard automated method (NeuroKit2), which showed MAEs from 22.29ms to 90.79ms.

Conclusions:

• QTNet demonstrates high accuracy and generalizability for QT interval estimation from single-lead ECGs.
• The model shows significant potential for automated, ubiquitous QT tracking in both clinical and ambulatory settings.
• QTNet facilitates continuous monitoring of patients at risk for QT prolongation, improving arrhythmia risk management.