Pain Classification Using EEG Evoked by Electrical Stimulation

Le résumé fourni par la source

Pain is a complex phenomenon to measure due to its composition of perceptual and affective processes, which is based on individual subjective nature.Healthcare needs a reliable pain quantification for pain scores and pain types, in which the former corresponds to the individual pain sensitivity and the latter corresponds to acute and chronic pain, resulting from the activation of nociceptive fibers, A and C, respectively.Recently, many research studies on using classification using features from cortical responses through electroencephalography (EEG) to classify pain perception levels besides using commercial pain measurement devices.However, the current classification systems, including the classifier model and feature, cannot achieve high accuracy for classifying multiple pain perception levels and nociceptive fibers activations.A major problem lies in the lack of effective features.To address this gap in research, we developed novel features with nonlinear analysis and Granger causality (GC) analysis for classifying multiple pain perception levels and activations of A-and C-fibers, respectively.The goal was to provide the effective features extracted from EEG induced by electrical stimulation for pain classification that would enable the prediction of pain levels and pain nerves activations.Moreover, aiming to demonstrate the possibility of EEG-based features in an online scenario.Several feature extraction approaches were proposed, including nonlinear analyses of Higuchi's fractal dimension, Grassberger-Procaccia correlation dimension, with functions of autocorrelation and moving variance for evaluating pain perception levels, and GC analysis for classifying nociceptive fibers activations.Furthermore, exploration of the different numbers of channels and trials for proof the concept of applying the current features for future online classification.The novelties and contributions were: 1) Using combined nonlinear features is effective to quantify pain into a maximum of four pain levels than using single nonlinear features; 2) The use of nonlinear feature extracted from EEG in the time domain with a number of trials less than 20 is not preferable because it cannot achieve sensible accuracy, which makes it difficult for instantaneous classification of pain perception levels; 3) For classification of Aand C-fibers activations, frequency-related GC has the best classification results among other GC features for the current data; 4) In a simulation of online analysis, apply detrending and dynamic time warping (DTW) to GC features extracted from EEG in the frequency domain can enhance the accuracy in the classification of nociceptive fibers activations.The obtained findings could be used further to improve the performances of the EEG-based pain classification system.I would like to acknowledge all the people who have supported me during the period of my Ph.D. program.First of all, I am exceptionally grateful to Yu sensei, my thesis advisor, for providing me the opportunity to do my Ph.D. at his lab and helping me to improve this work with his

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.