Morphometric similarity networks discriminate patients with lumbar disc herniation from healthy controls and predict pain intensity

Abstract

We used a recently advanced technique, morphometric similarity (MS), in a large sample of lumbar disc herniation patients with chronic pain (LDH-CP) to examine morphometric features derived from multimodal MRI data. To do so, we evenly allocated 136 LDH-CPs to exploratory and validation groups with matched healthy controls (HC), randomly chosen from the pool of 157 HCs. We developed three MS-based models to discriminate LDH-CPs from HCs and to predict the pain intensity of LDH-CPs. In addition, we created analogous models using resting state functional connectivity (FC) to perform the above discrimination and prediction of pain, in addition to comparing the performance of FC- and MS-based models and investigating if an ensemble model, combining morphometric features and resting-state signals, could improve performance. We conclude that 1) MS-based models were able to discriminate LDH-CPs from HCs and the MS networks (MSN) model performed best; 2) MSN was able to predict the pain intensity of LDH-CPs; 3) FC networks constructed were able to discriminate LDH-CPs from HCs, but they could not predict pain intensity; and 4) the ensemble model neither improved discrimination nor pain prediction performance. Generally, MSN is sensitive enough to uncover brain morphology alterations associated with chronic pain and provides novel insights regarding the neuropathology of chronic pain.

Keywords: chronic pain; lumbar disc herniation; morphometric similarity networks; pain intensity prediction; pain state discrimination.