Brain Mechanism of Acupuncture Treatment of Chronic Pain: An Individual-Level Positron Emission Tomography Study

Abstract

ObjectiveAcupuncture has been shown to be effective in the treatment of chronic pain. However, their neural mechanism underlying the effective acupuncture response to chronic pain is still unclear. We investigated whether metabolic patterns in the pain matrix network might predict acupuncture therapy responses in patients with primary dysmenorrhea (PDM) using a machine-learning-based multivariate pattern analysis (MVPA) on positron emission tomography data (PET).MethodsForty-two patients with PDM were selected and randomized into two groups: real acupuncture and sham acupuncture (three menstrual cycles). Brain metabolic data from the three special brain networks (the sensorimotor network (SMN), default mode network (DMN), and salience network (SN)) were extracted at the individual level by using PETSurfer in fluorine-18 fluorodeoxyglucose positron emission tomography (18F-FDG-PET) data. MVPA analysis based on metabolic network features was employed to predict the pain relief after treatment in the pooled group and real acupuncture treatment, separately.ResultsPaired t-tests revealed significant alterations in pain intensity after real but not sham acupuncture treatment. Traditional mass-univariate correlations between brain metabolic and alterations in pain intensity were not significant. The MVPA results showed that the brain metabolic pattern in the DMN and SMN did predict the pain relief in the pooled group of patients with PDM (R2 = 0.25, p = 0.005). In addition, the metabolic pattern in the DMN could predict the pain relief after treatment in the real acupuncture treatment group (R2 = 0.40, p = 0.01).ConclusionThis study indicates that the individual-level metabolic patterns in DMN is associated with real acupuncture treatment response in chronic pain. The present findings advanced the knowledge of the brain mechanism of the acupuncture treatment in chronic pain.

Progress of researches on machine learning combined with neuroimaging in the field of acupuncture

“…Liu et al [17] used the Least Absolute Shrinkage and Selection Algorithm (LASSO) to integrate sample salient features in order to construct support vector machine models. while Xu et al [27] showed a non-significant traditional mass univariate correlation between brain metabolism and altered pain intensity. Another study by Yu et al [28] applied the multivariate pattern analysis (MVPA) method of SVM to further optimize the prediction of the efficacy of acupuncture for primary dysmenorrhea (PDM) patients.③Most current predictions of acupuncture efficacy use subjective symptoms as labels, and thus the validity of the labels is influenced by individual cognitive biases, leading to heterogeneity of results.…”