Discrete Wavelet Transform based Pain Assessment using Multiple Physiological Signals

Over the past few years, there has been an increased interest in developing intelligent pain assessments based on physiological signals and Artificial Intelligence (AI). To provide patients with proper medical care, accurate measurement of pain assessment is crucial. The present study presents a time-frequency analysis-based pain assessment method based on multiple physiological signals and machine learning algorithms. In this study, EMG and ECG signals were collected from the Biovid Heat Physiological Signal Database (Part A). In this work, the HRV signals have been derived from ECG signals and used for pain assessment. In order to reduce the effects of noise (thermal and powerline) and other artifacts, these physiological signals are preprocessed through Butterworth Bandpass and a notch filter. Daubechies wavelets of different orders (1 to 8) are used to perform the time-frequency transformation, after which a set of statistical features, energy features, and higher-order statistical features are extracted. Finally, these features are fed into a K Nearest Neighbor (KNN) and a Regression Tree (RT) classifier to classify the pain levels. Based on the experimental results, the KNN classifier demonstrated superior accuracy in assessing no pain and high pain compared to the RT classifier. In addition, EMG signals provide more valuable information about pain compared to HRV signals. By using multiple physiological signals, the present study will contribute greatly to the design and development of a methodology for the assessment of pain.