Authors:
Elisavet Pavlidou
1
;
2
and
Manolis Tsiknakis
1
;
2
Affiliations:
1
Department of Electrical and Computer Engineering, Hellenic Mediterranean University, Estavromenos, Heraklion, 71410, Greece
;
2
Computational BioMedicine Laboratory, Institute of Computer Science, Foundation for Research & Technology-Hellas, Vassilika Vouton, Heraklion, 70013, Greece
Keyword(s):
Multimodal Pain Assessment, Pain Classification, Physiological Signals, Machine Learning, BioVid, ECG, GSR, EMG, Gender, Age.
Abstract:
Pain is a multidimensional and highly personalized sensation that affects individuals’ physical and emotional state. Visual analog scales, numeric rate indicators, and various questionnaires, all relying on patient-reported outcome measurements, are considered the “gold” standard methods for assessing the severity of pain. Nevertheless, self-report tools require cognitive, linguistic, and social abilities, which may manifest variations in certain populations such as neonates, individuals with intellectual disabilities, and those affected by dementia. The purpose of this study is to automate the process through multimodal physiological-data-driven machine-learning models in order to gain deeper insights into pain sensation. We developed a pipeline using electrocardiogram (ECG), galvanic skin response (GSR), and electromyogram (EMG), along with demographic information from the BioVid dataset. The Pan & Tompkins algorithm was applied for ECG signal processing, while statistical analysis
was used for feature extraction across all signals. Our study achieved 82.83% accuracy in the SVM classification task of baseline (BL) vs the highest level of pain (PA4) for females aged 20-35.
(More)