Deep Learning-Augmented ECG Analysis for Screening and Genotype Prediction of Congenital Long QT Syndrome

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Congenital long QT syndrome (LQTS) poses risks of syncope, arrhythmias, and sudden death.
• LQTS diagnosis is challenging as many patients present with normal or borderline QT intervals on ECG.
• Current diagnostic methods struggle to detect LQTS in a significant portion of affected individuals.

Purpose of the Study:

• To develop a deep learning neural network for identifying LQTS.
• To differentiate between LQTS genotypes (LQTS1 and LQTS2) using 12-lead ECG data.
• To compare the diagnostic accuracy of the AI model against traditional QTc interval measurements.

Main Methods:

• Utilized a deep learning convolutional neural network (CNN) model.
• Trained and validated the model on a large dataset of ECGs from the Hearts in Rhythm Organization Registry (HiRO).
• Compared CNN performance (AUC, F1 scores, sensitivity) with QTc interval-based detection in internal, external, and cross-sectional validation cohorts.

Main Results:

• The CNN demonstrated high diagnostic capacity for LQTS detection (AUC, 0.93) and genotype differentiation (AUC, 0.91) in external validation.
• The AI model significantly outperformed expert-measured QTc intervals in detecting LQTS, especially in patients with normal or borderline QTc intervals.
• CNN achieved superior F1 scores and sensitivity compared to QTc-based detection, indicating improved diagnostic accuracy.

Conclusions:

• Deep learning models can enhance the detection of congenital LQTS from resting ECGs.
• The developed CNN effectively differentiates between common LQTS genetic subtypes.
• Further validation in broader populations could support clinical application for suspected LQTS cases.