Pattern Classification Of Child Mental Diseases Based On Multimodal MRI

Abstract:Autism and ADHD are the common mental disorder diseases for the youth nowadays; these diseases have the characteristic of high incidence, subjective clinical diagnostic methods and have no good measure to cure. The magnetic resonance imaging (MRI) technology develops quickly in recent years, and has been an important tool to study the human brain, and promotes people to understand and research on the brain. MRI structure data can observe and measure the change of internal structure in brain visually; the functional MRI can reflect the brain function activities rapidly without damaging.We propose methods that can distinguish the autism and ADHD with the normal person automatically based on the multimodal MRI data. Then we contrast the difference between the patients and common people in the brain, extract and classify the features. Researching the classification method of this kind of disease has important implication not only can diagnose the patients early, also can get timely treatment and prognosis.In this paper, we have three datasets include:31autism patients,48ADHD patients and48control group. Then we preprocess the datasets and classify.1. The MRI structure data was preprocessed by the steps of tripping the skull and correcting bias field. After segmenting the three-dimensional brain image, we can obtain68components based on brain segments and five kinds of features, which comprise a total of340cortical features for each subject. Then statistic analysis and sequential forward selection (SFS) method were adopted to select the optimal features. At last using the support vector machine (SVM) to classify these optimal features. As a result, we achieved autism average prediction accuracies of93.67%for SVM classification.2. The functional MRI data was preprocessed by the steps head motion correction, image registration, image segmentation and so on. Then use the independent component analysis to decompose and get a series of statistical independent components, and extract the corresponding time series to build distance matrixs. Then maping the high-dimensional space of fMRI datas to a low dimensional space using the manifold analysis, and build corresponding spatial graph structures. At last classifying the features extracted by the graph, the result shows the classification accuracy of ADHD and control can reach87.6%, and the classification accuracy of autism and control nearly reaches80%.In conclusion, we realize there are differences between patients and control group through the analysis of the results of the pattern classification, and these differences can be used as distinct characteristics for identifying patients and normal people, at the same time we can get a good classification results. The research of this article can be built as the foundation for the clinical computer aided diagnosis system of autism and ADHD in future.