Arrhythmia Classification using Deep Learning and Machine Learning with Features Extracted from Waveform-based Signal...

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Arrhythmia poses a significant threat to cardiovascular health, necessitating early and accurate diagnosis.
• Current diagnostic methods for arrhythmia can be improved through advanced signal processing techniques.
• Automated classification of arrhythmia is crucial for timely clinical intervention.

Purpose of the Study:

• To introduce a novel waveform-based signal processing (WBSP) method for enhanced arrhythmia classification.
• To develop and evaluate machine learning (ML) and deep learning (DL) classifiers utilizing the WBSP method.
• To demonstrate the clinical relevance of WBSP in the early diagnosis of arrhythmia.

Main Methods:

• ECG signals were filtered, local minima identified, and baseline wander removed.
• Processed ECG signals were fitted with Gaussian functions to extract key parameters.
• ML-based and DL-based classifiers were developed and applied to the extracted waveform features.
• The MIT-BIH Arrhythmia Database was used for validation of the WBSP method.

Main Results:

• The WBSP method achieved state-of-the-art performance in arrhythmia classification.
• The best classifier demonstrated an overall accuracy of 98.8%.
• High sensitivities were achieved: 96.3% for class V and 98.6% for class Q, outperforming related works.
• Key waveform components (QRS similarity to Gaussian, QRS sharpness, P-wave duration/area) were identified as essential for classification.

Conclusions:

• The WBSP method offers a powerful tool for accurate and automated arrhythmia classification.
• The developed classifiers show significant potential for improving early arrhythmia diagnosis.
• This technique highlights the importance of specific ECG waveform components in clinical decision-making.