ECG signal classification based on deep CNN and BiLSTM.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Cardiovascular disease is a growing global health concern.
• Electrocardiograms (ECGs) are vital for heart health diagnosis.
• Limited medical resources necessitate advanced diagnostic tools like computer-aided systems.

Purpose of the Study:

• To develop an automated method for ECG identification and classification.
• To address the increasing demand for cardiovascular disease diagnosis with limited resources.

Main Methods:

• A novel deep learning model combining a 24-layer Deep Convolutional Neural Network (DCNN) and Bidirectional Long Short-Term Memory (BiLSTM).
• ECG signal preprocessing using wavelet transform and median filtering.
• A new loss function designed to stabilize training and improve model optimization.

Main Results:

• Achieved an accuracy rate of 89.3% and an F1 score of 0.891 on the 2017 PhysioNet/CINC challenge dataset.
• Demonstrated effective feature extraction using varied convolution kernel sizes.
• Validated through ten-fold cross-validation.

Conclusions:

• The proposed method offers improved accuracy in automatic ECG classification compared to existing approaches.
• This technique aids in clinical diagnosis and self-monitoring of conditions like atrial fibrillation.
• The study highlights the potential of deep learning in enhancing cardiovascular diagnostics.