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

In electrocardiograms (ECGs), the QT interval reflects the repolarization status of the heart. Currently, the QT interval automatically measured in 12-lead ECG is inaccurately derived as the median value. This approach fails to provide information on heterogeneous QT prolongation crucial for predicting arrhythmias, as it does not accurately represent the values of QT intervals in each lead.

Measurement of the true QT interval is challenging, even for EP cardiologists.

To address this challenge, we present a model that eliminates baseline wander and noise occurring during ECG and incorporates Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (Bi-LSTM) to estimate QT intervals.

The dataset used in this study were the QT Database, (two leads ECG data for 15 minutes from 63 patients) and Lobachevsky University database (12 leads ECG data for 10 seconds from 200 patients) with the annotation of P, QRS, and T waves. We structured these dataset into a ratio of 8:1:1 for train: Validation: Test. AI model combining CNN and Bi-LSTM was proposed to estimate the QT interval of ECG.

For the evaluation of deep learning, We also tested a similar recurrent neural network, GRU, with the same unit configuration. The results indicated that the method with two Bi-LSTM layers outperformed others in both Accuracy and F1-Score. For QT interval estimation in the QT database, Accuracy was 0.940 and F1-score was 0.938. Those of LU database were 0.927 and 0.921.

By incorporating CNN and Bi-LSTM, we conducted sufficiently accurate estimation of QT interval, leveraging temporal features of ECG. In our future research endeavours, we plan to deploy this model for predicting repolarization arrhythmias, specifically targeting conditions like long QT syndrome, drug-induced QT prolongation, and torsade pointes in ischemic cardiomyopathy.

Visualization of QT estimation with AI