| Major depressive disorder(MDD)is a common mental illness associated with persistent sadness and other psychiatric disorders.It has a high morbidity and is difficult to cure.It is a potential death factor that affects each and every family.Although the study of MDD has never stopped in recent years,the diagnosis about it has still not clarified.Neuroimaging is a type of clinical method for diagnosing mental illness.By using resting state functional magnetic resonance imaging(fMRI)to scan human brains,neuro images of brain can be got and further the lesions can be analyzed.On this basis,the brain can be divided into a number of regions which have relatively independent functions.Combined with the knowledge of complex network,these brain regions can be considered as nodes to build a scale of brain network.Extracting data from brain networks for classification can be used as a tool for neural imaging to diagnose mental illnesses.With the development of researching technology on brain networks,there are a lot of scholars who have applied it in the computer-aided diagnosis of MDD.At present,a large number of studies about classification on the brain network data of depression are based on a single spatial scale of the brain network,and the features used are mostly the clinical indicators or the basic building elements of brain networks.Some studies focus on the comparison of methods for feature selection and the choice of feature type,hoping to obtain the best solution for the diagnosis of MDD.Based on previous research,this thesis make a further discussion on the impact of the spatial scale of the brain network for the classification of depression.The main contents of this thesis are organized as follows:First,compare the classification of brain networks on different scales.The images of resting state fMRI are acquired from the patients with depression and normal subjects.After anatomical parcellation,five scales of brain networks with different number of nodes are obtained.The scales are 90,256,497,1003,1501 respectively.Extract local measures of brain network on different scales.Statistical analysis on the data of each type of subjects by using two-sample T-test.The measures with statistical significance were selected as the features of classifier.Make comparison after using the support vector machine(SVM)to calssify.Second,analysis the influence factors of the initial classification results.Firstly,in order to judge the influence of discriminative features,according to the brain network on different scales,replace the discriminative features from the brain network on large scale with the discriminative features from the brain network on small scale respectively,and combine with the discriminative features from the brain network on large scale to compare the result of classification.Secondly,in order to judge the influence of the number of features,extract the same number of discriminative features from the brain network on small scale as the number of features from the brain network on large scale to compare the result of classification.Third,is the smaller scale of networks better? In order to judge the influence of the classification results on different brain network scales,using the minimal redundancy maximal relevance(mRMR)to analysis the relevance of features to classification labels from brain networks on different scales and the redundancy between features from brain network on different scales.Fourthly,combine with the time complexity and the above analysis,the relative optimal classifier under the brain network on a certain scale is obtained,and the abnormal brain regions of the patients with depression is obtained.Considering these factors,classification accuracy increases along with the decrease of network scale.However,according to the analysis of this thesis,it is probably not better that the network scale is smaller.Using the brain network on 1003 scale to build the classifier for the classification of depression is better. |
PDF Full Text Request