Identifying resting state differences salient for resilience to chronic pain based on machine learning multivariate pattern analysis.

Abstract

Studies have documented behavior differences between more versus less resilient adults with chronic pain (CP), but the presence and nature of underlying neurophysiological differences have received scant attention. In this study, we attempted to identify regions of interest (ROIs) in which resting state (Rs) brain activity discriminated more from less resilient CP subgroups based on multiple kernel learning (MKL). More and less resilient community-dwellers with chronic musculoskeletal pain (70 women, 39 men) engaged in structural and functional magnetic resonance imaging (MRI) scans, wherein MKL assessed Rs activity based on amplitude of low frequency fluctuations (ALFF), fractional amplitudes of low frequency fluctuations (fALFF), and regional homogeneity (ReHo) modalities to identify ROIs most salient for discriminating more versus less resilient subgroups. Compared to classification based on single modalities, multi-modal classification based on combined fALFF and ReHo features achieved a substantially higher classification accuracy rate (79%). Brain regions with the best discriminative power included those implicated in pain processing, reward, executive function, goal-directed action, emotion regulation and resilience to mood disorders though variation trends were not consistent between more and less resilient subgroups. Results revealed patterns of Rs activity that serve as possible biomarkers for resilience to chronic musculoskeletal pain.