ADHD Classification Study Based On Resting State FMRI Data

Attention Deficit Hyperactivity Disorder(ADHD)is a common neuropsychiatric disorder,and the pathogenesis of ADHD is still unclear,so exploring the objective diagnostic criteria of ADHD is one of the important topics of current research.Resting-state functional magnetic resonance imaging(fMRI)data can reflect the spontaneous brain activity in resting state.By analyzing resting-state fMRI data,it is possible to find the signal differences between patients and the normal control group.According to the difference of signal and the machine learning algorithm,patients and normal control group can be accurately classified.Therefore,in recent years,the classification of ADHD based on resting state fMRI data has been gradually carriedout.In this study,resting-state fMRI data of 101 ADHD patients and 143 normal controls were studied.Firstly,the components of resting fMRI data were extracted,and the abnormal activity of components was analyzed.On this basis,two different models were constructed by combining feature extraction and pattern recognition methods to classify ADHD.Thesis research content is as follows:(1)ICA-SVM recognition model was constructed based on resting-statefMRI data,which is not easy to express mathematically.Independent Component Analysis(ICA)is suitable for multi-source signal analysis.ICA is used to extract components from resting-state fMRI data.The voxels containde in components are vectorized by parameters to construct feature vectors.Support Vector Machine(SVM)is selected as recognition network to construct ICA-SVM recognition model.The influence of component factors on the classification of ADHD patients and normal control group was analyzed.The experimental results showed that the average classification accuracy was71.92%.It proved that ICA-SVM recognition model was helpful to the classification of ADHD patients and normal control group.(2)In order to further improve the classification accuracy of ADHD patients and normal control group,this paper optimizes the recognition model based on ICA-SVM,and obtains DL-SVM recognition model.Because ICA-SVM recognition model is based on the sparse components of resting-state fMRI data,and the components extracted by ICA are less sparse,this paper sparse the extracted components by constructing a dictionary,and classifies them by using SVM based on sparse components,and constructs DL-SVM recognition model.