Does a classifier combining convolutional and recurrent neural networks improve the detection of myocardial infarction from single-lead ECG records compared to a pure convolutional neural network?
A combined convolutional and recurrent neural network classifier for single-lead wearable ECG devices achieves high diagnostic accuracy for detecting myocardial infarction.
Myocardial infarction (MI) is a medical emergency for which the early detection of symptoms is desirable. The prevalence of portable electrocardiogram (ECG) devices makes frequent screening for MI possible. In this study, we develop an MI classifier that combines both convolutional and recurrent neural networks, and is suitable for wearable ECG devices with only a single lead recording. It performs multiclass classification to discriminate the ECG records of MI from those of healthy individuals and patients with existing chronic heart conditions, as well as ECG records contaminated with noise. The method was tested on a dataset with MI ECG records and compared with a pure convolutional neural network and classifier with hand-crafted features. It was found that the addition of a recurrent layer improved the classification sensitivity by 28.0% compared to the convolutional neural network alone. Overall, it achieved 92.4% sensitivity, 97.7% specificity, a 97.2% positive predictive value, and a 94.6% F1 score.
Lui et al. (Mon,) studied this question.