Predicting pain location from resting-state brain fMRI

Abstract

Low back pain is a prevalent issue with few reliable treatments. Although there is great variation in clinical presentation within the low back pain population, little is known about the neurobiological mechanisms underlying these differences. In this study, we sought to stratify chronic low back pain patients (N = 275) into phenotypes characterized by correlated patterns of resting-state brain activity and sensory abnormalities (pain, numbness, and pins and needles) indicated on hand-drawn body maps. Our cross-decomposition analysis yielded phenotypes that resemble previously documented mechanistic pain types, revealing distinct brain connectivity patterns associated with different clinical presentations. Our model was then used to predict pain body maps from fMRI data in a small novel dataset of chronic pain subjects, suggesting that these relationships may generalize to other chronic pain conditions. Our results support the utility of resting-state fMRI in understanding the heterogeneity of chronic pain, which may be leveraged to develop more targeted pain treatments.