Supervised learning methods for pathological arterial pulse wave differentiation: A SVM and neural networks approach

Area of Science:

• Biomedical Engineering
• Machine Learning in Healthcare
• Signal Processing

Background:

• Arterial pulse wave (APW) analysis is crucial for cardiovascular disease diagnosis.
• Distinguishing healthy, pathologic, and noisy APW signals is challenging in clinical settings.
• Novel optical systems offer new avenues for APW data acquisition.

Purpose of the Study:

• To develop an automated supervised learning method for classifying APW signals.
• To differentiate between healthy, pathologic, and noisy APW waveforms.
• To reduce diagnostic bias in cardiovascular disease assessment using APW.

Main Methods:

• Utilized a dataset of APW signals from 213 subjects (healthy and non-healthy).
• Extracted 39 pulse features including morphologic, statistical, cross-correlation, and wavelet features.
• Employed Support Vector Machine Recursive Feature Elimination (SVM RFE) for feature selection and compared Support Vector Machine (SVM) with Artificial Neural Network (ANN) classifiers.

Main Results:

• SVM achieved a statistically superior performance with an average accuracy of 0.9917 and F-Measure of 0.9925.
• ANN achieved an average accuracy of 0.9847 and F-Measure of 0.9852.
• Performance varied significantly with the number of features used in the SVM classifier.

Conclusions:

• SVM demonstrates higher performance than ANN for APW classification.
• The proposed automated method shows potential for differentiating healthy, pathologic, and noise APW signals.
• This approach can aid in reducing bias in clinical diagnosis of cardiovascular diseases via APW analysis.