Identifying neuropathic pain using 18F-FDG micro-PET: A multivariate pattern analysis

- We acquired FDG micro-PET images in the brain of awake rats with neuropathic pain.
- Multivariate approach identified neuropathic pain animals with high accuracy.
- Brain regions yielding high prediction performance were investigated.
- Univariate approach resulted in reduced performance of prediction.

Pain is a multidimensional experience emerging from the flow of information between multiple brain regions. A growing body of evidence suggests that pathological pain causes plastic changes of various brain regions. Here, we hypothesized that the induction of neuropathic pain alters distributed patterns of the resting-state brain activity in animal models, and capturing the altered pattern would enable identification of neuropathic pain at the individual level. We acquired micro-positron emission tomography with [18F]fluorodeoxyglucose (FDG micro-PET) images in awake rats with spinal nerve ligation (SNL) and without (sham) (SNL group, n = 13; sham group, n = 10). Multivariate pattern analysis (MVPA) with linear support vector machine (SVM) successfully identified the brain with SNL (92.31% sensitivity, 90.00% specificity, and 91.30% total accuracy). Predictive brain regions with increased metabolism were mainly located in prefrontal-limbic-brainstem areas including the anterior olfactory nucleus (AON), insular cortex (IC), piriform cortex (PC), septal area (SA), basal forebrain/preoptic area (BF/POA), amygdala (AMY), hypothalamus (HT), rostral ventromedial medulla (RVM) and the ventral midbrain (VMB). In contrast, predictive regions with decreased metabolism were observed in widespread cortical areas including secondary somatosensory cortex (S2), occipital cortex (OC), temporal cortex (TC), retrosplenial cortex (RSC), and the cerebellum (CBL). We also applied the univariate approach and obtained reduced prediction performance compared to MVPA. Our results suggest that developing neuroimaging-based diagnostic tools for pathological pain can be achieved by considering patterns of the resting-state brain activity.