Functional Connectomes Based On Different Brain Atlas In The Identification Of Primary Trigeminal Neuralgia Patients: A Preliminary Study

Objective:To explore the difference in brain functional connectivity patterns between primary trigeminal neuralgia(PTN)patients and healthy controls(HC)by using support vector machine(SVM)for machine learning analysis,which was trying to build a neuroimaging model to identify PTN patients,so as to provide a imaging basis for clinical diagnosis.Methods:Resting-state functional magnetic resonance imaging(rs-f MRI)scans were performed on 31 unilateral PTN patients and 31 HC.After the preprocessing of the rs-f MRI data,the functional connectomes were constructed according to the AAL90 brain atlas,AAL116 brain atlas,Fan246 brain atlas,and Shen268 brain atlas,respectively,which were the feature variables that the machine learning analysis needs.Feature selection was performed by the method of T-test,then the model was trained and tested with the selected data by using the SVM algorithm and leave-one-out cross validation method until all the participants are treated as test set once,and the average classification accuracy of the test set was used as the classification performance of the model after LOOCV procedure 62 iterations.Next,apply the permutation test to assess the reliability of the model.Finally,compare the classification effect of different brain atlas,select the brain atlas with the best classification performance,find the characteristics that survived after each LOOCV and calculate its weights to reflect the contributions among classification.Result:(1)The accuracy of the SVM classification models,based on AAL90 brain atlas,AAL116 brain atlas,Fan246 brain atlas,and Shen268 brain atlas,all exceeded80%,and the P-values after 5,000 permutation tests were all less than 0.001.Among them,the Fan246 brain atlas had the highest accuracy,up to 87.10%.(2)The features predicting individual differences were not located in specific brain regions but distributed throughout the brain,but functional connectivity between frontal and cortical and subcortical was the main factor in classification.Besides,the top 20 nodes in the number of connections were located in the bilateral prefrontal lobe,left middle temporal gyrus,right inferior parietal lobule,and left precentral gyrus.Conclusions:(1)Machine learning methods can effectively distinguish PTN patients from HC,and the functional connectomes features may be potential biomarkers of PTN.(2)The choice of brain atlas has some impact on the performance of the model.(3)The SVM model constructed by the Fan246 brain atlas highlights the significant role of functional connectivity between frontal and cortical and subcortical in the classification.At the same time,it highlights the role of sensory discrimination components,cognitive and attention components and motor components of pain information in the course of the disease course.