Research On Feature Analysis Method Of Neurological Diseases Based On Automatic Annotation Fiber Clustering

The white matter fiber tractography based on diffusion magnetic resonance imaging(d MRI)is an important method for non-invasively studying neurological diseases.At present,in studies of brain anatomy changes,voxel-based atlas is widely used to first identify the cortical area,then compare the difference of diffusion metrics in the anatomical structure of the brain between groups.However,this method is challenged by cortical variability or prematurely terminating tractography,and investigating the whole white matter tract may weaken the original change in specific sections and results in statistical characteristics that are not obvious.In order to overcome these shortcomings,this paper proposes a feature analysis method of neurological diseases based on automatic annotation fiber clustering(AAFC),which is applied to investigate PD and SWEDD in subdivisions of white matter tracts.The main work and results are as follows:(1)Aiming at the shortcomings of traditional neurological disease analysis methods,this paper proposes a feature analysis method of neurological diseases based on AAFC.AAFC is first applied to segment tractography of HCP data into 800 fiber clusters,each cluster can be assigned to an anatomical tract annotation according to the fiber geometry and cortical terminations.An automatic fiber quantization algorithm is then used to calculate the diffusion properties along the center of the fiber cluster trajectory.The diffusion properties are constantly changing along the trajectory of the fiber cluster,thus the analysis with the regional mean metrics may weaken the characteristics.The benefits of the method are that it sets up a tract-based model that can be highly robust in identifying corresponding white matter structures across subjects without requiring individual anatomical prior information,and provides a fine brain parcellation,which allows for investigation of neurological disease in subdivisions of white matter tracts.(2)Aiming at the problem that the similar clinical symptoms of PD and SWEDD make it difficult to accurately diagnose two diseases,this paper applies the feature analysis method to extract the anatomical features along each cluster defined in cingulum bundle(CB),thalamo-frontal(TF),thalamo-parietal(TP),thalamo-occipital(TO),corpus callosum(CC)in PD,SWEDD and normal controls(NC),which are fed into a Support Vector Machine(SVM)classifier framework to separate the three groups.In comparison with NC,PD shows significant difference in two clusters in TF,one cluster in TP,and one cluster in TO,while SWEDD shows no significant difference.Three clusters in CB commonly exhibit significant difference in PD versus SWEDD and NC versus SWEDD.These results suggest PD presents more significant effect on thalamo tracts than SWEDD,and there are unique microstructural changes in CB tract in SWEDD.In classification experiments,anatomical features of the test set data are extracted in the same way and used to distinguish the three groups.By using features in subdivisions of white matter tracts,the accuracy of PD-NC,PD-SWEDD,SWEDDNC is 80%,75%,and 72.5%,all of which are higher than that at the tract level.This proves differences in subdivisions of white matter tracts are more sensitive in the anatomical changes of PD and SWEDD,which is of great significance for accurate diagnosis of two types of diseases.