At present 51 million 100 thousand people in the world have been affected by the Conduct Disorder. As people pay more and more attention to the human brain, functional magnetic resonance imaging (fMRI) technology contributes greatly to the process of human brain exploration. The abstract human brain can be used in data form for the researchers to explore the truth behind the data. The secret becomes very meaningful. This study classifies fMRI data of patients with conduct disorder to provide reference for clinical research.
The experiment used the SIEMENS 3-TMRI scanner to get the resting state function MRI images of the subjects, and use the SPM8 and DPARSF software based on the MATLAB platform to produce the time series of each person’s brain area. The Pearson correlation coefficient of the time series between each two brain regions is calculated and the brain function connection and characteristics are established. The feature is sorted and selected by the F-Score method. Then the feature data is classified by SVM.
The learning model is established by using the radial basis kernel function and the mesh parameter optimization in the classification process. Finally, the learning model is established. The classifier is evaluated using a cross validation method. The experimental results show that the number of different characteristics has an impact on the accuracy of the classification of the conduct disorder. The classification accuracy can be reached when the number of characteristics of the most regional division can be reached to 94.44%.
With the rapid development of society and the increasing pressure in life, mental disorders have become people’s nightmares, such as depression, conduct disorder(CD), attention deficit hyperactivity disorder and so on.
CD of child and adolescent period is a common mental Disorder, is a continuous, repeatedly infringe others’ basic rights and violation of social norms school-age problem. For this problem, there is no perfect treatment, magnetic resonance imaging (MRI) with no damage to the advantages of the structure and function of the human brain, and is widely use in the research of brain science,Currently, the fMRI research on CD mainly focuses on three main aspects: executive function, emotional regulation and control, and decision-making mechanism.
In the study, the rsfMRI(Resting state fMRI) data were used to calculate the Pearson correlation coefficient between each two brain regions as the classification feature. In order to facilitate classification, these features were sorted by F score, and then the classification model was trained with these significant features. At last, the performance of the classifier was evaluated with the accuracy obtained by cross validation.
Datasets. This study was approved by the Biomedical Ethics Committee of Second XiangYa Hospital of Central South University of China.Eighteen CD right-handed patients (ages 15-17, male) were recruited from the youth detention center in China’s HuNan province. Exclusion criteria included being younger than 15 and older than 18, having a history of other physical diseases, emotional disorders and neurological diseases.From Changsha city of HuNan province at the same time, the community and local school recruited 18 age, sex, and matches by the education level of healthy subjects (healthy development committee (TD), the ages of 15 and 17 years old, male) took part in the study, as a control group, all the subjects and their parents or caregivers have signed a written agreement, guarantee the stability of the experimental process.According to the clinical interview of Chinese edition k-sads -PL (status quo and survival period of school-age children’s affective disorder and schizophrenia) [6-8], doctors and interviewees confirmed that the current control group did not have any neurological disease and other psychiatric history.In addition, only CD participants meeting the k-sads-pl standard were included in the CD group.All participants were also able to assess the occurrence of depression and anxiety symptoms in the previous week by using the Birleson depression self-rating scale (DSRS) and the child anxiety-related mood disorder screening table (SCARED) .
Feature extraction and selection. Average return for the whole brain problems haven’t solve, this topic not to return, to calculate each time series between the two brain regions of Pearson correlation coefficient can get a whole brain function connection, the 4005 correlation coefficient of each subjects as the brain connections, as the characteristics of each sample.
In order to reduce the training time and redundant features and optimize the training model, corresponding feature selection methods should be used for feature reduction and feature extraction.Therefore, in order to improve classification performance,F score method was selected in this experiment to extract features. After calculating the F-score of the characteristic i of a sample, the F-score is sorted. The first 250 F-scores corresponding to the order from large to small are extracted from the reference experiment for the first time, and then the steps of feature selection are completed.
Classification. In this study, we used matlab function to carry out SVM classification. Different kernel functions can generate different support vector machine models, which can not only reduce the impact caused by dimensions but also improve the efficiency of calculation. The commonly used kernel functions include Linear Function,Polynomial Function,Radial Basis Function,Sigmoid Function. The Radial Basis Function is adopted here.
In the process of using Leave-One-Out cross-validation, the classification is correctly recorded as 1, and incorrectly recorded as 0. After the 36 groups of samples are tested, the total accuracy of this feature is obtained by dividing the 36 groups of samples by the total number of samples. As shown in Figure 1, the broken line graph on the number and accuracy of brain functional connectivity is:
It can be seen from figure 1 that the accuracy rate is as high as 94.44% when the number of features is selected as 400. But the limitation of this study is that the data volume is too small, and the results are compared without using other classification methods. The two aspects will be improved in the future.