Automatic detection of arrhythmia from imbalanced ECG database using CNN model with SMOTE

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Cardiovascular diseases pose a significant health risk, with timely diagnosis crucial for patient survival.
• Cardiac arrhythmia detection from electrocardiogram (ECG) signals is challenging due to subtle signal variations.
• Existing methods often require complex preprocessing steps like denoising and QRS complex segmentation.

Purpose of the Study:

• To develop and evaluate an 11-layer deep convolutional neural network (CNN) for automated cardiac arrhythmia classification.
• To classify the MIT-BIH arrhythmia database into five standard classes without prior signal denoising or QRS complex detection.
• To address class imbalance issues in cardiac arrhythmia datasets using the SMOTE technique.

Main Methods:

• An 11-layer deep convolutional neural network (CNN) architecture was designed for end-to-end ECG signal classification.
• The MIT-BIH arrhythmia database was artificially oversampled using the SMOTE technique to mitigate class imbalance.
• The CNN model was trained on the augmented dataset and validated on the original dataset, bypassing denoising and QRS segmentation.

Main Results:

• The developed CNN model demonstrated superior performance compared to existing literature methods.
• The model achieved high precision, recall, and F-score for arrhythmia classification.
• An optimal accuracy of 98.30% was recorded using a 70:30 train-test data split.

Conclusions:

• The proposed deep CNN model offers an effective and simplified approach for cardiac arrhythmia detection.
• The methodology eliminates the need for ECG signal denoising and QRS complex segmentation, streamlining the diagnostic process.
• This advanced CNN model shows significant potential for improving computer-aided diagnosis systems in cardiology.