Automatic Detection of Short-Term Atrial Fibrillation Segments Based on Frequency Slice Wavelet Transform and Machine...

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Atrial fibrillation (AF) is a common cardiac arrhythmia linked to serious cardiovascular and cerebrovascular conditions.
• Accurate automatic detection of AF from electrocardiogram (ECG) signals is crucial for patient management.
• Traditional methods focusing solely on atrial activity are insufficient; incorporating ventricular activity analysis is key.

Purpose of the Study:

• To develop and validate a robust method for automatic AF detection using ECG signals.
• To leverage time-frequency analysis of P-QRS-T waveforms for improved AF identification.
• To compare the efficacy of various machine learning classifiers for AF detection.

Main Methods:

• Utilized frequency slice wavelet transform (FSWT) for time-frequency analysis of short-term ECG segments.
• Applied an average sliding window to convert 2D time-frequency matrices into 1D feature vectors.
• Employed five machine learning (ML) techniques, including Gaussian-kernel support vector machine (GKSVM) with Bayesian optimization, for classification.

Main Results:

• The FSWT-GKSVM model achieved 100% accuracy on the training set and 93.4% on the validation set.
• The model demonstrated high performance on an unseen test set with 98.15% accuracy, 96.43% sensitivity, and 100% specificity.
• The proposed model exhibited superior stability and robustness compared to existing methods.

Conclusions:

• The FSWT-GKSVM model offers a stable and robust approach for the automatic detection of atrial fibrillation.
• This method effectively utilizes both atrial and ventricular activity information from ECG signals.
• The findings support the clinical application of this advanced signal processing and ML technique for AF diagnosis.