Subthreshold Depression Classification Based On VMF Directional Statistical Model And DTI Data

Subthreshold depression is likely to develop into severe depression.At present,there are few studies on subthreshold depression in the field of brain imaging neurobiology.In the study of white matter disease classification,the traditional methods are mostly based on the classification of multi-modal medical image data or the scalar information of diffusion tensor imaging(DTI)data.These method does not fully exploit the advantages of DTI data implied in white matter research.The use of DTI to study and classify the abnormal white matter in subthreshold depressive people is beneficial to the exploration of abnormal white matter in subthreshold depression and the prediction of the disease.Based on the research of traditional scalar feature index classification methods,this paper proposes a new statistical method of direction information statistics of the VMF direction statistical model for the study of DTI white matter disease classification.According to the tensor data,this method can describe the characteristics of the fiber orientation information,fit the fiber direction information to the unit hypersphere,and do direction information statistics.Firstly,the expectation-maximization algorithm is used to estimate the mixed model parameters based on the tensor data.Then the directional data statistics are made according to the mixed model parameters,and the sum of the included angles between the main feature directions and the aggregation degree k around the feature vector are fitted as directional feature,finally apply the directional feature to study white matter classification.In order to verify the effectiveness of the proposed method in the study of white matter disease.This paper selects the DTI data of subthreshold depression as the object of feature classification study.Firstly,the ROI method is combined with FSL and DTIStudio magnetic resonance image processing software package to preprocess these DTI data,including generating scalar images,image registration,and brain regions traditional feature index extraction.According to the brain area index analysis,the left corticospinal tract DTI data was selected as the simulation data of the VMF direction statistical methodproposed in this paper.The result show that the accuracy of classification based on traditional scalar features reaches 67%,and the accuracy of classification using new directional information indicators reaches 73%.Combining the new directional information indicators with traditional scalar feature indicators can achieve a classification accuracy of 79%,which is 12% higher than the traditional feature classification.The directional feature extraction method in this paper makes use of the directional information of tensor data,which supplements the deficiencies of scalar features in traditional white matter classification studies and improves the classification accuracy.