Research On Classification And Prediction Of ADHD Based On Resting-state FMRI

ADHD is a very common psychiatric disorder in childhood. The pathogenesis of ADHD is not clear, and the diagnosis of ADHD depends mainly on subjective assessment based on DSM-4. Objective diagnosis and effective treatment of ADHD is one of the important topics in the field of neuroscience. In recent years, the application of resting state fMRI technique in the pathological analysis of neuropsychiatric disorders, including ADHD, demonstrated outstanding advantages. Studies on ADHD classification based on resting state fMRI are gradually expanded.In this study, ADHD classification and prediction were carried out based on resting state fMRI data of 97 ADHD patients and 121 normal controls. Specifically, independent functional networks were firstly extracted based on the resting state fMRI data, analyses of brain functional network abnormalities in patients with ADHD were then performed, and ADHD classification and prediction were finally carried out using SVM and random forest. Details are as follows:First, the analysis of abnormal independent networks of ADHD patients.Based resting state fMRI data, we extracted the functionally independent networks of each sample, and then compared the brain functional networks of ADHD patients and normal controls. Results show that ADHD patients were abnormal in networks including the right frontoparietal network, the posterior default network, the orbitofrontal network, motor control loops and the auditory-motor network.Secondly, the classification and prediction of ADHD based on brain functional independent networks features and SVM. Based on the foregoing results of statistical analysis, we conducted univariate feature selection. On this basis, classification and prediction models were built using support vector machines with leave-one-out cross-validation. Results show that the classification rate of ADHD based on spatial maps of brain functional independent networks is 73.90%, while on the ADHD prediction, the correlation coefficient of predicted scores and actual scores up to 0.33.Finally, the ADHD classification and prediction based on brain functional independent networks features and random forests.Independently brain functional network map features were selected based on univariate method, using random forest method conducted ADHD classification and prediction, classification accuracy and prediction accuracy has greatly improved:the highest classification rate up to 78.00%, while the correlation coefficient of ADHD predicted scores and actual scores can also be reached 0.51.ADHD-200 data released so far, based on resting state fMRI studies of ADHD classification in this field is mainly based on fMRI signal regional homogeneity and multi-ROI functional connectivity features.In this study, we first attempt to conduct ADHD classification and prediction research based on resting state fMRI extracted brain functional independent networks features and achieved considerable international coverage classification accuracy. In addition, we first attempt to carry out the classification of mental disorders based on independent brain network temporal features, experimental results are not satisfactory, the results prompted us that lower temporal features dimension of brain functional independent networks can provide extremely limited information, is not conducive to the classification of neuropsychiatric disorders.