Research On The Mechanism Of Acupuncture Brain Effect After Hemiplegia Due To Stroke Based On Multimodal Magnetic Resonance Imaging And Machine Learnin

Background:Hemiplegia is the main sequelae of stroke,and the motor rehabilitation of patients has multiple medical-psychological-social significance.Acupuncture,as one of the characteristic therapies of traditional Chinese medicine,has a positive effect on the rehabilitation of stroke hemiplegia.With the development of neuroimaging,multimodal magnetic resonance imaging(MRI)technology and machine learning(ML)have provided new research directions for brain function remodeling and the central effect mechanism of acupuncture after stroke.Objective:This study intended to classify and predict wheather there was the minimal clinically significant differences in patients’ motor recovery after stroke hemiplegia by ML and multimodal MRI,and initially explore potential neuroimaging features of the brain regions related to stroke hemiplegia recovery.Combined with static functional connectivity(FC)and dynamic co-activation patterns(CAP),the multi-dimensional visualization of brain plasticity and immediate effects of acupuncture after stroke hemiplegia could be identified,to further reveal the temporal and spatial characteristics of brain functional remodeling and acupuncture central mechanism after stroke hemiplegia.Methods:This study was divided into two parts.1 In the first section,69 patients with unilateral motor pathway(basal ganglia and/or radial coronal region)injury after cerebral infarction were included.On the day of enrollment,functional magnetic resonance imaging(fMRI)were performed,and a 2-week follow-up was conducted.The Fugl-Meyer Assessment(FMA)was used to evaluate the motor impairment and recovery.According to the improvement of motor function(△ FMA)around 2 weeks,after removing the ceiling effect,49 patients who were actually included in the analysis were divided into a minimal clinically important difference(MCID)group(28 patients)and a non-minimal clinically important difference(N-MCID)group(21 patients).The patients’brain function indicators(Degree Centrality,DC)and structural indicators(Fractional Anisotropy,FA)were used as neuroimaging features to predict the classification of two groups through ML which was performed by PRoNTo software.We aimed to identifiy the feature brain regions of interest(ROIs)and explore potential neuroimaging biomarkers for motor recovery in patients with stroke hemiplegia.2 In the second section,a total of 26 patients in the MCID group,15 patients in the N-MCID group and 26 healthy control(HC)matched the age and gender were included,and all the subjects underwent MRI scanning while acupuncture at Yanglingquan(GB34)acupoint.The feature brain regions with high predictive weights obtained in the first section were selected as the region of interest(ROI)to compare and analyze the differences in functional connectivity(FC)and co-activation pattern(CAP)between patients and HC,patients with different recovery outcomes under resting and acupuncture states.We proposed to explore the spatiotemporal characteristics of brain function remodeling after stroke hemiplegia and the response patterns of the brain under the acupuncture state,and try to elucidate the possible central effect mechanism of acupuncture on brain function remodeling after stroke hemiplegia.Results:1 In terms of classification prediction:Functional indexes(DC)had a total accuracy of 75.51%,a sensitivity of 82.14%and a specificity of 66.67%(P<0.05),showing good prediction performance(area under curve of 0.80).The feature brain regions mainly involved the contralesional caudate nucleus and putamen nucleus(ROI 1),the ipsilesional supplementary motor areas(SMA)and superior frontal gyrus(ROI 2),the contralesional precentral gyrus,paracentral lobule,SMA,dorsolateral superior frontal gyrus and the postcentral gyrus(ROI 3),as well as psychological and cognitive-related brain regions such as the frontotemporal lobe,orbital gyrus and occipital lobe.The structural index(FA)had a total accuracy of 63.69%(P<0.05),a sensitivity of 75.00%(P>0.05)and a specificity of 67.74%(P>0.05),showing a relatively general predictive performance(area under curve of 0.63).The feature brain regions mainly involve the bilateral superior cerebellar peduncle(ROI 1,ROI 3)and the partial regions of limbic system(fornix,cingulate gyrus).2 In terms of functional connectivity:2.1 In the resting state,compared with HC,the FC between ROI 1 and the bilateral cerebellum,middle cingulate gyrus,paracentral lobule and the ipsilesional SMA was reduced in the MCID.The FC between ROI 2 and the bilateral thalamus,caudate nucleus,middle cingulate gyrus,SMA,the ipsilesional precentral gyrus,postcentral gyrus,rolandic operculum,inferior frontal gyrus,opercular part,temporal pole:superior temporal gyrus,superior temporal gyrus,and the contralesional insula,rolandic operculum were reduced.The FC between ROI 3 and the ipsilesional superior parietal gyrus,pstcentral gyrus,precuneus,precentral gyrus,middle frontal gyrus,caudate nucleus,thalamus,rolandic operculum and opercular part of inferior frontal gyrus were reduced(GRF correction,voxel threshold of P<0.01,cluster threshold of P<0.05).In the N-MCID,there were decreased FCs between ROI 1 and the contralesional triangular part of inferior frontal gyrus,inferior frontal gyrus pars orbitalis,cerebellum,decreased FCs between ROI 2 and the bilateral anterior and middle cingulate gyrus,the contralesional supramarginal gyrus,middle temporal gyrus,and decrease FCs between ROI 3 and the ipsilesional caudate nucleus and thalamus compared with HC(GRF correction,voxel threshold of P<0.01,cluster threshold of P<0.05).2.2 In the acupuncture state,compared with HC,the FC between ROI 1 and the bilateral cerebellum,the contralesional middle occipital gyrus,middle temporal gyrus and angular gyrus was reduced,the FC between ROI 2 and the ipsilesional pallidus,putamen,insula,hippocampus,amygdala,caudate nucleus,the bilateral anterior and medial cingulate SMA,and the FC between ROI 3 and the ipsilesional precentral gyrus,postcentral gyrus,the contralesional middle occipital gyrus,inferior occipital gyrus were decreased(GRF correction,voxel threshold of P<0.01,cluster threshold of P<0.05).Compared with the N-MCID,the FC between ROI 1 and the ipsilesional lingual gyrus,cerebellum,calcarine fissure and surrounding cortex,cuneate and inferior occipital gyrus were reduced in the MCID(GRF correction,voxel threshold of P<0.01,cluster threshold of P<0.05).There were no statistically significant differences of FC between the N-MCID and HC(GRF correction,voxel threshold of P>0.01,cluster threshold of P>0.05).3 In terms of co-activation pattern:3.1 In the resting state,a total of 6 co-activation patterns were identified based on the BOLD signals of ROI 1 and ROI 2,and 9 co-activation patterns were identified based on the BOLD signals of ROI 3.The identified CAPs were not a single activation state,showing a co-existence of positive and negative activations as a whole,but each with a bias.The spatial distribution of these activated brain regions overlaped with the layout of the brain networks,such as the default mode network(DMN),which may be related to the functional connectivity between main brain regions in the brain network.By compared and analyzed the differences in dynamic characteristics of CAPs among different groups and their correlation with ΔFMA,it was found that among the CAPs identified by each ROI,some CAPs appeared less frequently,with fewer occurrences in the N-MCID and HC,while the MCID showed more activity,which is likely to be the unique co-activation patterns of the MCID and also be positively correlated with the motor recovery.The dynamic characteristics of the N-MCID in some CAPs are significantly lower than those of the MCID or HC,indicating that dynamism was suppressed.3.2 In the acupuncture state,a total of 9 co-activation patterns were identified based on the BOLD signals of ROI 1 and ROI 3,and 11 co-activation patterns were identified based on the BOLD signals of ROI 2.Further comparisons were made on the dynamic characteristics of CAPs among different groups.Compared with the CAPs of resting state,there were some CAPs of acupuncture state were similar,and also some new CAPs were different.Acupuncture at Yanglingquan enhanced the activation of the positive activation regions in the CAP that originally existed in the resting state,reduced the range of the negative activation regions and weakened the activation.In the acupuncture state,the newly emerged CAP had both positive and negative activation,with more positive activation of the brain regions.Acupuncture not only increased the number of CAPs,but also adjusted the spatial consistency between CAPs.In addition,some brain regions in DMN exhibited activity(either positive or negative)in both resting and acupuncture states,so DMN may be the main response target for functional activation.By comparing and analyzing the differences in dynamic characteristics of CAPs among different groups in the acupuncture state and their correlation with ΔFMA,it was also found that the dynamic characteristics of the MCID showed more differences than those of HC,and were positively correlated with ΔFMA.These differences were mainly concentrated in newly appeared or infrequently appeared CAP,which may be the unique acupuncture effect patterns of the MCID.Conclusion:The functional and structural indexes of neuroimaging after stroke hemiplegia could predict the outcomes of motor recovery,and the functional indicators had better sensitivity and specificity.The functional feature brain regions with predictive attributes may be used as potential neuroimaging biomarkers for stroke hemiplegia rehabilitation.In patients with stroke hemiplegia,the resting FC was abnormally reduced.Acupuncture on Yanglingquan might adjust the abnormal FC between the bilateral cerebral hemispheres,so that it tended to a healthy state.Brain function remodeling after stroke was a complex multi-regional dynamic interaction process,with changing temporal and spatial characteristics,acupuncture could induce and regulate the dynamic co-activation patterns to promote the rehabilitation of stroke hemiplegia.