ECG beat classification based on discrete wavelet transformation and nearest neighbour classifier

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Myocardial Infarction (MI) is a critical condition caused by reduced blood supply to the heart muscle.
• Classifying specific MI types, like Anteroseptal MI (ASMI), is crucial for timely and accurate diagnosis.
• Electrocardiogram (ECG) signals contain vital information for diagnosing cardiac events.

Purpose of the Study:

• To propose a statistical classification method for Anteroseptal Myocardial Infarction (ASMI).
• To evaluate the effectiveness of different distance metrics in an ECG-based classification system.
• To enhance the accuracy of automatic ASMI detection from ECG data.

Main Methods:

• Utilized multi-resolution wavelet analysis and thresholding for noise elimination and feature extraction from ECG signals.
• Implemented a Nearest Neighbour (NN) classification rule using temporal and amplitude features from chest leads (v1-v4).
• Compared Euclidean and Mahalanobis distances as metrics for the NN classifier.

Main Results:

• The Mahalanobis distance-based NN rule achieved a classification accuracy of 95.14%.
• The Euclidean distance-based NN rule yielded a classification accuracy of 81.83%.
• Mahalanobis distance proved superior to Euclidean distance for ASMI classification in this study.

Conclusions:

• The proposed statistical approach, particularly with Mahalanobis distance, offers a highly accurate method for automated ASMI diagnosis.
• Wavelet analysis and NN classification are effective tools for analyzing ECG signals in cardiology.
• This method provides a valuable tool for improving the diagnostic capabilities for Anteroseptal Myocardial Infarction.