O-WCNN: an optimized integration of spatial and spectral feature map for arrhythmia classification

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Arrhythmia diagnosis is crucial for reducing cardiovascular disease mortality.
• Traditional methods require improvement in accuracy and efficiency.

Purpose of the Study:

• To propose a novel Convolutional Neural Network (CNN) model for enhanced arrhythmia classification.
• To improve the efficiency and accuracy of arrhythmia detection using advanced deep learning techniques.

Main Methods:

• Developed a multi-channel CNN model concatenating spectral and spatial features.
• Utilized depthwise separable convolutions for parameter efficiency.
• Optimized hyperparameters using the Hyperopt library (SMBO algorithm).
• Evaluated performance with tenfold cross-validation (inter-patient and intra-patient protocols).

Main Results:

• Achieved 99.48% accuracy for ventricular ectopic beat (VEB) classification.
• Achieved 99.46% accuracy for supraventricular ectopic beat (SVEB) classification.
• Demonstrated significant performance improvement over state-of-the-art models.

Conclusions:

• The proposed CNN model offers superior accuracy and efficiency for arrhythmia classification.
• This advancement can aid in reducing mortality from cardiovascular diseases.
• The model's architecture and optimization strategy represent a significant step forward in cardiac monitoring technology.