Artificial intelligence-assisted QT interval measurement in 12-lead ECGs

The QT interval is crucial for assessing cardiac repolarization. However, accurately measuring it remains challenging, as conventional methods using median values from 12-lead ECGs often fail to detect heterogeneous QT prolongation—an essential factor in arrhythmia risk prediction.

Our goal was to develop an AI model utilizing CNN, Bi-LSTM networks, and ResNet to enhance the accuracy of QT interval estimation after ECG signal preprocessing.

We used two datasets: QT Database (QTDB) – 15-minute, two-lead ECGs from 63 patients and Lobachevsky University Database (LUDB) – 10-second, 12-lead ECGs from 143 patients, annotated with P, QRS, and T waves. To mitigate signal artifacts, we systematically removed 1,000 samples from both the initial and terminal segments of each recording. Additional noise was filtered using the Neurokit2 library, yielding cleaner ECG signals.

We then developed a ResNet18-LSTM deep learning model to predict QT intervals automatically. Here is a visualization of QT interval measurements using our model.

Performance evaluation (accuracy, precision, recall, and F1-score) yielded. AI model's showed the a ccuracy 0.940, F1-score 0.938 at QTDB and the Accuracy 0.927, F1-score 0.921 at LUDB, respectively.

Both datasets demonstrated high performance, with F1-scores exceeding 0.9.

Conculsion: This study successfully reduced ECG noise, improving QT interval measurement accuracy through deep learning. Our AI model with CNN, Bi-LSTM and ResNet model providing precise QT estimations. In the future, we aim to deploy this model for predicting repolarization-related arrhythmias, particularly in long QT syndrome, drug-induced QT prolongation, and ischemic cardiomyopathy.

AI model