Altered habenular connectivity in chronic low back pain: An fMRI and machine learning study

Abstract

The habenula has been implicated in the pathogenesis of pain and analgesia, while evidence concerning its function in chronic low back pain (cLBP) is sparse. This study aims to investigate the resting-state functional connectivity (rsFC) and effective connectivity of the habenula in 52 patients with cLBP and 52 healthy controls (HCs) and assess the feasibility of distinguishing cLBP from HCs based on connectivity by machine learning methods. Our results indicated significantly enhanced rsFC of the habenula-left superior frontal cortex (SFC), habenula-right thalamus, and habenula-bilateral insular pathways as well as decreased rsFC of the habenula-pons pathway in cLBP patients compared to HCs. Dynamic causal modelling revealed significantly enhanced effective connectivity from the right thalamus to right habenula in cLBP patients compared with HCs. RsFC of the habenula-SFC was positively correlated with pain intensities and Hamilton Depression scores in the cLBP group. RsFC of the habenula-right insula was negatively correlated with pain duration in the cLBP group. Additionally, the combination of the rsFC of the habenula-SFC, habenula-thalamus, and habenula-pons pathways could reliably distinguish cLBP patients from HCs with an accuracy of 75.9% by support vector machine, which was validated in an independent cohort (N = 68, accuracy = 68.8%, p = .001). Linear regression and random forest could also distinguish cLBP and HCs in the independent cohort (accuracy = 73.9 and 55.9%, respectively). Overall, these findings provide evidence that cLBP may be associated with abnormal rsFC and effective connectivity of the habenula, and highlight the promise of machine learning in chronic pain discrimination.

Keywords: chronic low back pain; dynamic causal modelling; habenula; resting-state functional connectivity; support vector machine.

FIGURE 1

Location of the habenula in the brain. The left and right habenulas are adjacent to the medial dorsal nucleus of the thalamus. The red‐yellow part in the upper row is corresponding to the grey‐white ones in the lower row. The brighter regions represent the larger possibility of the habenula.

FIGURE 2

Resting‐state functional connectivity of the habenula. There was significantly enhanced (red) and decreased (blue) resting‐state functional connectivity of the habenula in patients with chronic low back pain compared to healthy controls. The connectivity values of the habenula with the thalamus and pons were obtained from small‐volume correction analyses. SFC, superior frontal cortex

FIGURE 3

Effective connectivity of the habenula. (a) Locations of brain regions used in the dynamic causal modelling. (b) Seven models used in the DCM analysis. (c) Results from Bayesian model selection. BMS, Bayesian model selection; CLBP, chronic low back pain; DCM, dynamic causal modelling; FFX, fixed effect; LHb, left habenula; pon, pons; RFX, random effect; RHb, right habenula; SFC, superior prefrontal cortex; Tha, thalamus

FIGURE 4

Winning models in dynamic causal modelling (DCM) at the group level. The numbers represent the connectivity parameters (Hz) of the winning model in the patients with chronic low back pain and healthy controls. The solid lines and dotted lines represent connectivity values greater/less than 0.1 Hz, respectively. The red octagons represent significant between‐group differences in that pathway. CLBP, chronic low back pain; HC, healthy controls; LHb, left habenula; pon, pons; RHb, right habenula; SFC, superior prefrontal cortex; Tha, thalamus

FIGURE 5

Machine learning model performance of different features. The accuracy (ACC) and area under the curve (AUC) of the three features in the test cohort are shown. The horizontal axis represents 10 splits, and the vertical axis represents the ACC and/or AUC values. The average ACC/AUC of an experiment is marked with dotted lines of the same colour as the broken line listed in the lower right corner. ACC, accuracy; AUC, area under the curve

FIGURE 6

Receiver operating characteristic curves in machine learning. The receiver operating characteristic curves demonstrated the classification performance of combining the resting‐state functional connectivity of the habenula with all three brain regions by three machine learning methods. AUC, area under the curve; FPR, false‐positive rate; LR, linear regression; RF, random forest; SVM, support vector machine; TPR, true positive rate; train: the training set (N = 104); test, the test set (N = 68).