基于静息态fMRI的三叉神经痛模式分类研究

Application of Functional Magnetic Resonance Imaging in Trigeminal Neuralgia

目的采用机器学习技术进行三叉神经痛患者的模式分类研究,探讨基于术前静息态功能磁共振(resting state-functional magnetic resonance imaging,rs-fMRI)数据预测患者术后疼痛及面部麻木程度的可行性。方法采集34名三叉神经痛患者半月节射频热凝术(percutaneous radiofrequency thermocoagulation,PRT)术前的fMRI数据,并提取全脑功能连接及ReHo值,作为机器学习算法的训练数据。采集患者术后6个月的VAS评分及面部麻木程度评分,将患者分别按照疼痛及麻木程度分为轻度/中重度两类,构建支持向量机(support vector machine,SVM)分类器,用训练数据对分类器进行训练,采用留一交叉验证法检验分类器的泛化能力,采用置换检验验证分类器的可靠性。结果 SVM分类器对术后6个月疼痛程度预测的准确率为82.35%,面部麻木程度预测的准确率为73.53%。结论以三叉神经痛患者静息态全脑功能连接和局部一致性为特征数据,用于预测PRT术后疼痛缓解程度及面部麻木程度具有较高的准确度,具备操作上的可行性和一定的临床实用价值。

Objective To explore the feasibility to predict the outcome of pain release and facial numbness in TN pa- tients after PRT procedure by using the pre-surgery resting-state fMRI data.Methods Thirty-four TN patients under- went the resting-state fMRI scanning one week before the PRT procedure.The whole brain functional connectivity (FC)and ReHo values calculated from fMRI data were used as training feature set.The VAS scores and degree of fa- cial numbness were measured 6months after PRT procedure.To classify the patients a support vector machine classifier along with principal component analysis was used to solve the classification problem.Leave-one-out cross-validation strategy was applied to estimate the generalization ability of the classifier.The permutation tests were also used to as- sess the classifier performance.Results The correct classification rate of the no pain and painful patients 6months after PRT procedure was 82.35%,and the prediction accuracy of patients with mild facial numbness or moderate-severe fa- cial numbness was 73.53%.Conclusion We have validated the hypothesis that pre-surgery fMRI data correlate with pain release and facial numbness of TN patients who received PRT procedure.Multivariate decoding of fMRI data with SVM enables a robust prediction of TN patients'medium-term outcomes.