Brain network characteristics of pain: a network-based statistics and graph theory-based EEG analysis in healthy individuals and chronic pain patients

Traditional pain neuroimaging research has predominantly focused on regional brain activation, while emerging evidence indicates that pain perception and chronification are mediated by dynamic interactions across large-scale brain networks. Electroencephalography (EEG) is ideal for studying these interactions, yet existing studies show inconsistent findings due to lack of methodological rigour and unified framework linking experimental to clinical pain. This thesis employs network-based statistics and graph theory to characterise the brain network organisation of pain and further validates translational applicability via machine learning. Using the debiased weighted phase lag index (dwPLI, a functional connectivity measure resistant to volume conduction artefacts) alongside multi-layer graph construction, this thesis addresses methodological discrepancies (absolute vs. relative comparisons; eyes-open/EO vs. eyes-closed/EC states) and investigates alpha-band reorganisation. Three studies are presented: (1) tonic experimental pain in healthy volunteers; (2) short-term neuroplasticity reflected by resting-state reconfigurations pre-/post-sensory stimulation; and (3) resting-state EEG dynamics between chronic pain (CP) patients and healthy controls (HC). Global network inferences (GNIs) revealed a continuum of pain-related network reorganisation. Tonic pain shifted brain networks from segregation to adaptive integration, whereas post-stimulation rest facilitated segregation, reflecting adaptive neural plasticity. In contrast, CP was characterised by a maladaptive signature of increased integration and diminished small-worldness, which was driven primarily by the chronic back pain sub-group. Crucially, this core signature was consistent across EO and EC states. The support vector machine classifier demonstrated the high diagnostic potential of GNIs, with classification performance yielding AUC-ROC values of 0.94 (tonic pain), 0.88 (pre-/post-stimulation), and 0.91 (CP vs. HC). Methodologically, integrating absolute and relative comparison methods with multi-modal EO/EC protocols was found to optimise biomarker detection. Overall, this thesis fosters the methodological identification and validation of GNI-based biomarkers that bridge experimental and clinical pain, establishing a network-based framework for understanding pain chronification and providing a foundation for objective subtype-specific pain diagnostics.