An automatic method for arterial pulse waveform recognition using KNN and SVM classifiers.

Area of Science:

• Biomedical Engineering
• Cardiovascular Health
• Signal Processing

Background:

• Arterial pulse waveform (APW) analysis is crucial for cardiovascular risk assessment.
• Non-contact optical sensors offer a promising method for APW measurement, particularly at the carotid artery.
• Automated signal processing is needed to effectively analyze APW data.

Purpose of the Study:

• To develop and compare automatic methods for extracting and classifying APW signals and noise.
• To investigate a wide range of features for APW signal characterization.
• To identify the most relevant features for accurate signal-noise discrimination.

Main Methods:

• Implemented and compared k-nearest neighbours and support vector machine (SVM) classifiers.
• Extracted a comprehensive set of 37 features from APW signals, categorized into amplitude, time, wavelet, cross-correlation, and frequency domains.
• Utilized support vector machine recursive feature elimination (SVM-RFE) for optimal feature selection.

Main Results:

• The SVM classifier demonstrated superior performance in discriminating between APW signals and noise.
• An optimal subset of features was identified through SVM-RFE, enhancing classification accuracy.
• The best achieved accuracy for signal-noise discrimination was 0.952.

Conclusions:

• The proposed automatic method, particularly with the SVM classifier and optimized feature subset, is effective for APW signal analysis.
• This approach holds potential for improving non-contact cardiovascular risk assessment using optical sensor data.
• The comprehensive feature engineering and selection process provides valuable insights for APW signal processing.