A Parallel Cross Convolutional Recurrent Neural Network for Automatic Imbalanced ECG Arrhythmia Detection with...

Area of Science:

• Cardiology
• Artificial Intelligence
• Signal Processing

Background:

• Cardiovascular diseases (CVD) are a leading cause of global mortality.
• Automatic arrhythmia detection using electrocardiogram (ECG) and deep learning (DL) is crucial for early diagnosis and treatment.
• Existing DL models struggle with multi-class imbalanced ECG data, limiting their performance.

Purpose of the Study:

• To propose a novel parallel cross convolutional recurrent neural network (CNN) for improved arrhythmia detection in imbalanced ECG signals.
• To address the challenge of multi-class imbalanced data in ECG arrhythmia detection.
• To enhance the classification performance of DL models for CVD diagnosis.

Main Methods:

• Developed a novel parallel cross convolutional recurrent neural network integrating recurrent neural networks and 2D CNNs.
• Utilized continuous wavelet transform (CWT) to convert ECG signals into 2D time-frequency scalograms.
• Employed 2D CNNs to learn spatial information from scalograms and recurrent networks for temporal characteristics.

Main Results:

• The proposed model effectively learns features from imbalanced ECG samples.
• Achieved significant improvements in model convergence and classification accuracy.
• Demonstrated superior performance compared to existing hierarchical network models on the MIT-BIH arrhythmia dataset.

Conclusions:

• The novel parallel cross convolutional recurrent neural network is highly effective for arrhythmia detection.
• The model successfully handles imbalanced ECG data, improving diagnostic accuracy for CVD.
• This approach offers a significant advancement over previous methods for automated ECG analysis.