QTcNet: a deep learning model for direct heart rate corrected QT interval estimation.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Automated electrocardiogram (ECG) interpretation, particularly QTc interval measurement, often shows discrepancies compared to expert clinical assessments.
• Existing automated methods struggle with accuracy, necessitating improved algorithms for reliable cardiovascular risk stratification.

Purpose of the Study:

• To develop and validate QTcNet, a deep learning model designed to enhance the accuracy of automated QTc interval measurements from ECG data.
• To compare the performance of QTcNet against conventional algorithms and expert readings across multiple independent datasets.

Main Methods:

• A regression-based convolutional neural network (QTcNet) was developed and trained on a large dataset of algorithm-labeled ECGs (120,300), including internal hospital data and MIMIC-IV.
• The model underwent evaluation on three independent datasets (PTB, QTcMS, ECGRDVQ), with performance assessed against expert QTc measurements.
• Model explainability was analyzed using Integrated Gradient maps to ensure physiological relevance.

Main Results:

• QTcNet demonstrated a significant reduction in mean absolute error (MAE) from 23.4 ms to 13.4 ms and root mean square error (RMSE) from 40.1 ms to 22.1 ms.
• The model nearly halved the number of large outliers (>50 ms) in QTc measurements.
• Explainability analysis confirmed QTcNet's focus on physiologically relevant ECG components (QRS onset, T wave offset).

Conclusions:

• QTcNet, trained on large-scale data, consistently surpasses traditional algorithms in QTc measurement accuracy across external validation sets.
• While fine-tuning can adapt the model to specific cohorts, it may reduce generalizability.
• The QTcNet model and code are publicly released to encourage further research and development.