Research On Pattern Recognition Methods Based On Brain Magnetic Resonance Imaging Data

The human brain has more than hundreds of billions of connections,and is ”one of the most complex systems in the universe.” The brain produces all the human cognition,emotion,behavior,and thought,and it can be said that the brain shapes and makes human mankind itself.The emergence of magnetic resonance imaging(MRI)provides a powerful tool for us to explore the mysteries of the brain.Structural MRI(structural MRI,sMRI)and functional MRI(functional MRI,fMRI)images are of great significance in studying the structure of the brain and decoding the activation level and cognitive state of brain function.In recent years,univariate and multivariate pattern analysis methods have often been used in brain image analysis.With the increasing attention of multi-modal,multicenter and multi-group data,compared with the traditional single-task learning method,the multi-task feature learning method has gradually been paid attention to.This paper uses univariable and multivariate learning in the single-task method,to accomplish the pattern classification and abnormal feature learning between two groups of subjects;and proposes a multi-task learning framework for multi-center disease data and multi-group data classification.The paper mainly includes the following four aspects of work:Multivariate classification of nasopharyngeal carcinoma patients before and after radiotherapy.In the second chapter,the two groups of patients were classified by studying the abnormal patterns of whole brain functional connectivity in patients with nasopharyngeal carcinoma caused by radiotherapy.We collected the two groups of fMRI images of the subjects,defined 160 region of interest(ROI)in the whole brain,and generated the whole brain functional connectivity map of each subject.We used the multivariate pattern analysis method to classify the two groups,and got promising classification accuracy,and found the most distinguishing different functional connections between the two groups.At the same time,the correlation between these different functional connectivity and cognitive score was assessed.The results showed that the two groups of patients had45 abnormal functional connectivity,which are mainly located in the connection between the three brain networks,including cerebellum,sensorimotor and cingulo-opercular networks.There were significant correlations between five abnormal functional connectivity and cognitive score,especially the attention score.These results suggest that radiotherapy not only causes impairment of motor function,but also leads to cognitive dysfunction,especially attention deficit.These abnormal patterns of functional connectivity may serve as potential biomarkers and provide valuable targets for further functional recovery therapies.This study demonstrates the effectiveness of the multivariate pattern recognition method in classifying two groups of subjects,and its ability in accurately identifying the patterns of whole brain functional connectivity differences between the two groups.Univariate learning on studying the abnormal patterns of cerebellar functional connectivity in patients with nasopharyngeal carcinoma caused by radiotherapy.The third chapter focuses on the different patterns of cerebellar-cerebral functional connectivity between the patients with nasopharyngeal carcinoma receiving radiotherapy and without radiotherapy.We selected 10 symmetrical ROIs in the cerebellum,representing 5different functional networks,and generated the cerebellar-cerebral brain functional connectivity maps of all subjects.The functional connectivity between the cerebellum and the brain was analyzed in two groups using a rigorous univariate statistical test.Three significant functional connections were found,and the three connections were also found to be significantly associated with scores on the cognitive scale.We hypothesized that radiotherapy significantly altered these three cerebellar brain functional connections,and that the three abnormal connections may be associated with attention,memory,and executive dysfunction in patients undergoing radiotherapy.These significant changes in the functional connectivity and functional network of the cerebellum may serve as potential biomarkers for further understanding of brain cognitive abnormalities induced by radiotherapy.This study shows that univariate analysis is simple and effective in extracting the differences between two groups of subjects.Proposing a multi-task learning framework in classifying schizophrenia subjects from three centers.With the arrival of the big data era,the research of multi-center data becomes very important.However,data heterogeneity was brought by data acquisition with different machines,which poses a great challenge in multi center fusion learning.In this chapter,we focus on the classification of schizophrenia data collecting from multiple-centers.The goal is to improve the accuracy of multi-center classification over single-center classification.We collected sMRI data from three groups of schizophrenia and normal people at three research centers and extracted the gray matter volume information of each subject.With the assumption that different centers of disease data share the same abnormalities associated with the disease,and also the center-specific features,we propose a multi task learning framework to learn the site-shared and site-specific features simultaneously,and highlights the site-shared feature weights which can reflect the pathological mechanism of the disease to construct the classifier.The results show that the classification accuracy of multi task learning is significantly higher than the accuracy of classification based on the single-center,and the accuracy of the three center merging into a large data set.Multi-task learning method also are able to obtain consistent features of multi center,which have been widely reported in previous research on schizophrenia.Our research shows that the framework of multi-task learning provides an effective pattern classification method for multi-center fusion learning problem,and has great value in dealing with data heterogeneity.Proposing a multi-task learning framework for the classification multi-group of subjects including schizophrenia patients before and after treatment and normal people.The analysis and recognition of schizophrenia patients before and after treatment,and normal people can help to make use of the information between the three groups of data,and improve the classification performance of each classification task.In the fifth chapter,the fMRI images of schizophrenia patients before treatment and after treatment and normal subjects were collected to generate the whole brain functional connectivity patterns of each subject.Four different tasks were defined,including normal subjects verse all patients,normal subjects verse untreated patients,normal subjects verse treated patients,and untreated verse treated patients.We propose a multi-task learning method for feature learning and classification of the four classification tasks simultaneously.The results show that the performance of multi-task learning is better than that of single-task learning.Multi-task learning can accurately learn the significantly different features of each classification task,and these features provide imaging evidence for understanding the differences between subjects in different groups.This study shows the advantages of multi-task learning in multi-group data classification and feature extraction.