A Signal Processing Framework for the Detection of Abnormal Cardiac Episodes

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Long-term electrocardiogram (ECG) recordings are crucial for diagnosing abnormal cardiac episodes.
• Manual analysis of these recordings is time-consuming and labor-intensive for cardiologists.

Purpose of the Study:

• To develop and validate an automated signal processing framework for detecting abnormal cardiac episodes in long-term ECG signals.
• To optimize an algorithm that reduces the manual burden in cardiac diagnostics.

Main Methods:

• ECG signals were pre-processed using basis pursuit sparsely decomposed tunable-Q wavelet transform (BPSD-TQWT) to remove noise.
• 44 time, frequency, and time-frequency domain features were extracted.
• Support Vector Machine (SVM), K-nearest neighbour (KNN), and other classifiers were evaluated for performance.

Main Results:

• The Support Vector Machine (SVM) model demonstrated superior performance among the tested classifiers.
• SVM achieved high accuracy in detecting abnormal episodes across both open-source and proprietary databases, with results up to 99.89% accuracy.
• Cross-database validation showed robust performance, indicating generalizability of the proposed framework.

Conclusions:

• The proposed automated framework effectively detects abnormal cardiac episodes in long-term ECG recordings.
• The high performance suggests its potential application in autonomous diagnostic systems for cardiology.
• This approach can significantly aid cardiologists by reducing analysis time and improving diagnostic efficiency.