Key result
Hybrid CNN-LSTM model classifies short-duration ECG rhythms with ~84% accuracy.
Why the study?
Automated detection of atrial fibrillation using deep learning techniques on very short ECG recordings had not been performed with high accuracy before.
Absolute Event Rate: 83.5% vs 63.7%
Deep learning techniques, specifically a CNN-LSTM hybrid model, can automatically detect atrial fibrillation from short-duration ECG recordings with 83.5% accuracy without requiring manual feature extraction.
May aid preliminary ECG screening; leaves open prospective validation before clinical use.
Cardiac arrhythmia occurs when electric impulses co-ordinating the heart beat malfunction. In atrial fibrillation (AF), heart beats irregularly, which increases the risk of stroke and heart diseases. In this paper, the authors apply deep learning techniques to categorize ECG to classes of normal, AF or others. They utilize convolutional neural network (CNN) and hybrid of CNN and other deep learning architectures of long short-term memory (LSTM), gated recurrent unit (GRU) and recurrent neural network (RNN) to automatically detect AF. No feature extraction/selection is required. A number of trials of experiments were run (1000 epochs) to arrive at the optimum value of parameters. Learning rate was fixed in the range [0.01 0.5]. A high accuracy of 83.5% is obtained using separate training and testing datasets in classifying the input ECG as belonging to normal, abnormal (atrial fibrillation) and others with CNN-LSTM. This is the first work to perform classification mainly to detect AF using ECG recordings of very small duration (average 30s) with high accuracy employing deep learning techniques.
No takes yet. Share an insight, caveat, or question.
G et al. (2020) studied Atrial Fibrillation (n=8,528). CNN-LSTM deep learning architecture vs. CNN 7-layer architecture was evaluated on Accuracy of AF detection. A hybrid CNN-LSTM deep learning architecture achieved an accuracy of 83.5% in classifying short-duration ECG recordings as normal, atrial fibrillation, or other rhythms.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: