Han, Yiyuan (2024) Assessing Pain from Neurophysiological Signals: Machine Learning Approaches Using Functional Connectivity. Doctoral thesis, University of Essex.
Han, Yiyuan (2024) Assessing Pain from Neurophysiological Signals: Machine Learning Approaches Using Functional Connectivity. Doctoral thesis, University of Essex.
Han, Yiyuan (2024) Assessing Pain from Neurophysiological Signals: Machine Learning Approaches Using Functional Connectivity. Doctoral thesis, University of Essex.
Abstract
The assessment of pain remains a significant challenge in healthcare, which always relies on subjective measures based on behaviour. However, the unresponsive patients, e.g., the ones with the disorder of consciousness, cannot self-report pain. Addressing this challenge, this thesis develops machine learning approaches for pain assessment using electroencephalography (EEG). This research initially quests the suitable neural biomarkers of pain, which firstly analysed the neural biomarkers of integration. By examining functional connectivity and cross-frequency coupling, phase-based functional connectivity from the alpha band emerged as a robust feature, excelling in both prediction accuracy and computational efficiency. Consequently, it was employed in subsequent model development. Considering the individual variability in pain processing, this study adopted cross-domain transfer learning strategies, targeting unresponsive patients as they cannot provide training labels. Inspired by the contributions of deep learning models to transfer learning, a convolutional neural network (CNN) was implemented. The CNN model demonstrated good performances in both pain prediction and subject recognition with the subject involved in model training, which suggested its potential role in cross-subject transfer learning. Furthermore, the study figured out the activation patterns of functional connectivity produced by the CNN using Gradient-weighted Class Activation Mapping (Grad-CAM). The patterns revealed the critical involvement of frontal, parietal, and occipital brain regions in functional connectivity specific to pain assessment, in line with established physiological findings. Nevertheless, the work also identified the challenge of isolating pain-specific features due to the extensive distribution of functional connectivity contributing to subject recognition. Overall, this thesis emphasises the potential of phase-based functional connectivity from the alpha band as an ideal neural biomarker for pain and individual differences. It also introduces CNN-based transfer learning as a promising avenue for cross-subject pain assessment.
Item Type: | Thesis (Doctoral) |
---|---|
Uncontrolled Keywords: | pain, machine learning, functional connectivity, transfer learning |
Subjects: | Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QP Physiology |
Divisions: | Faculty of Science and Health > Computer Science and Electronic Engineering, School of |
Depositing User: | Yiyuan Han |
Date Deposited: | 12 Jun 2024 15:28 |
Last Modified: | 12 Jun 2024 15:28 |
URI: | http://repository.essex.ac.uk/id/eprint/38532 |
Available files
Filename: YiyuanHan_Thesis_CSEE.pdf