Key result
Transformer and LSTM models achieved higher sensitivity (0.97, 0.98) and specificity (0.99) for QRS detection in tachycardia ECGs compared to a traditional wavelet algorithm (0.67, 0.71).
Why the study?
Deep learning models for QRS detection have primarily focused on normal heart rates, leaving their accuracy on fast-beating ECGs (heart rate > 100) unclear.
Do Transformer and LSTM deep learning models improve QRS detection accuracy compared to traditional wavelet algorithms in fast-beating ECGs?
Do Transformer and LSTM deep learning models improve QRS detection accuracy compared to traditional wavelet algorithms in fast-beating ECGs?
Deep learning models (Transformer and LSTM) significantly outperform traditional wavelet algorithms in detecting QRS complexes in fast-beating ECGs (tachycardia).
Answers are grounded in this paper and cite their sources.
May improve QRS detection in tachycardia ECGs; hypothesis-generating and requires prospective validation before clinical use.
The QRS complex is one of the most distinctive features of ECG signals, and it can play an essential role in understanding heart rate (HR) and identifying heart diseases. Although many deep learning models applied to QRS detection have shown good performances, they are limited to ECGs with normal heart rates. The accuracies of deep learning models on the detection of QRS for fast beating ECGs whose HR > 100 are not elucidated. The present study aims to develop ECG segmentation models based on Transformer and LSTM to locate QRS and compare the detection performance between traditional algorithms and deep learning models. Firstly, we collected 267 clinically used 12-lead ECGs whose QRS are pre-annotated by cardiologists as a training dataset. Secondly, an ECG window with 250 points in length is adopted to be the input of this developed Transformer and LSTM models, and the ECG windows are augmented and then generate 1,124,604 windows. The results indicate that the sensitivities and specificities of the QRS detection by Transformer, LSTM, and traditional wavelet algorithm are 0.97/0.99, 0.98/0.99, and 0.67/0.71, respectively. In conclusion, both deep learning models, such as Transformer and LSTM have better accuracies in the detection of QRS beats in ECG tachycardia, which can be applied to clinical applications in the near future.
No takes yet. Share an insight, caveat, or question.
Hong et al. (2022) studied Tachycardia (n=267). Transformer and LSTM deep learning models vs. Traditional wavelet algorithm was evaluated on QRS detection sensitivity. Transformer and LSTM models achieved higher sensitivity (0.97, 0.98) and specificity (0.99) for QRS detection in tachycardia ECGs compared to a traditional wavelet algorithm (0.67, 0.71).
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: