Feature Extraction From Brain Magnetic Resonance Images

The study of the structure and function of cerebral cortex is a hot topic in brain science.Thanks to the emergence of magnetic resonance imaging technology,it is possible to extract features from structural and functional brain images.In this paper,based on the point cloud atlas of the cerebral cortex obtained by segmenting the structural magnetic resonance images,a new point cloud extraction algorithm was firstly proposed to investigate the differences in the brain structure between male and female,as well as the differences in the structure between healthy people and schizophrenics.Secondly,for functional magnetic resonance images,the hemispheric specialization of functional connectivity patterns in schizophrenics with auditory and non-auditory hallucination was further investigated by calculating the autonomy index.The main work of this paper is as the follows:A new local morphological feature description method based on cortical vertex cloud is proposed.In this paper,through preprocessing the international open available dataset,we obtained the data of point cloud of cerebral cortex.By adjusting the parameters of the new point cloud feature extraction algorithm and choosing the appropriate numbers of key points and feature dimensions,the accuracy of gender classification can reach 91.33%.The results showed that the new method achieved good performance on gender classification.Because the new method is not affected by the parameters of magnetic resonance scanner,the new feature description method provides a good solution for the analysis of multi-center brain imaging data.The application of cortical vertex cloud description method in multi-center brain imaging data.In this paper,the new algorithm is applied to the classification of multi-center gender data and multi-center schizophrenic data.By calculating the point cloud data of structural images on nine centers,the effect of the new method on multi-center gender or schizophrenic classification was investigated.The results showed that the brain structure of male and female,as well as schizophrenia and healthy people were separable,and the method was feasible in migration classification of multi-center brain imaging data.The application of hemispheric specialization feature description method in schizophrenia.This method is a feature description method for functional magnetic resonance imaging.By calculating hemispheric autonomy indices of each voxel in the cerebral cortex,we investigate the specific specialization patterns of auditory and non-auditory verbal hallucinations in schizophrenia.The results showed that the patients with and without hallucinations could be clustered into two groups with an accuracy of80.6%,and the hemispheric specialization of left temporal areas increased in patients with hallucinations,while the symptoms of hallucinations may disappear once the left frontal regions emerge increased specialization.The contributions of this paper are listed as follows: 1)The point cloud feature extraction method is applied to brain imaging data.Then we applied this method to the study of gender and clinical medicine classification and solved the problem of multi-center analysis of brain images from the perspective of traditional machine learning methods;2)The hemispheric specialization feature description method was applied to schizophrenia data,and the hemispheric specialization of functional connectivity between auditory and non-auditory verbal hallucinations was explored for the first time.This index has good clustering effect and provided a potential biomarker to differentiate schizophrenic patients with and without auditory verbal hallucinations.