Morphological Analysis And Classification Of Subcortical Structures Based On Magnetic Resonance Imaging

The brain is the most complex and fully functional organ in humans,it can both generate thought and control consciousness.With the popularization of structural magnetic resonance imaging(sMRI)technology,morphological analysis and classification of brain diseases with the help of sMRI has become a research focus in recent years.As a routine item for clinical examination,sMRI has the advantages of low acquisition cost and good reliability of retesting,so it has a very high application prospect in the field of pathological analysis and computer-aided diagnosis.However,existing mainstream morphological analysis features,such as those based on volume and surface area,are often not sensitive enough to local morphological changes of the subcortical structure,and lack the ability to capture early subtle morphological changes or development patterns.Therefore,how to better capture the subtle cortical structure morphological changes and extract valuable features for pathological classification and diagnosis has become an important issue to be solved urgently.Through the introduction of novel high-sensitivity surface morphological features and the designing of novel classification pipeline,this thesis has the following four contributions:First,by introducing several novel morphometric features,the subregional deformation analysis of the two subcortical structures — the hippocampus and amygdala of the major depressive disorder(MDD)group was performed.There are 4 main findings in this study: ○1 the significant morphological changes in the hippocampus and amygdala of the MDD group were found mainly occurred in specific subregions of the corresponding structure rather than the global of them;○2 for the MDD group receiving antidepressant treatment,the atrophy and expansion mechanisms of the hippocampus and amygdala are coexisted;○3 a positive linear correlation was found between the degree of subregional atrophy in the hippocampus and amygdala and the severity of depression;○4 after receiving antidepressant treatment,the hippocampus and amygdala have stronger morphological elasticity and have a tendency to gradually approach helthy control(HC)subjects in morphology(confirming the morphological efficacy of antidepressants).Second,the above morphometric features were used to explore the morphological changes of six subcortical structures — the nucleus accumbens,amygdala,caudate nucleus,hippocampus,pallidum and putamen in male patients with autism spectrum disorder(ASD).We have two main findings: ○1 evaluating the morphological changes of the subcortical structures in male ASD patients from the perspective of volume alone may lead to heterogeneous conclusions;○2 the morphological elastic effects caused by drug treatment were not extensive existed in all subcortical structures of male ASD patients,they were only effective for specific subregions of specific structures.Third,we applied the spectral matching ideas derived from Laplace-Beltrami operator to the morphological study of the paired subcortical structures of male ASD patients,and carried out research from the relatively novel perspective of brain laterality.Our analysis based on Laplace eigenvalues revealed that male ASD patients had significant morphologically lateral changes in specific but not all subcortical structures compared to typical development(TD)male subjects.This discovery based on a public database and large dataset provided a new research breakthrough for the development of lateral and symmetric patterns in the brains of male ASD patients.At the same time,our results also showed that there existed significant difference between the lateral patterns of the subcortical structures of female ASD patients and the male ASD patients(the lateral patterns of the subcortical structures of female ASD patients were more similar to that of female TD subjects),therefore from the morphometric perspective provided an explanation for the pathological differences between male and female patients with ASD.Fourth,based on the high-dimensional morphometric features of the subcortical structure,we designed a novel and universal classification pipeline.This classification pipeline includes surface-based feature generation(guaranteeing feature sensitivity),dimensionality reduction schemes based on sparse coding and dictionary learning(maintaining geometric spatial patterns of morphometric features),Max-pooling(further dimension reduction),and adaptive-based optimizer’s ensemble classifier selection(pick up the best of them).This pipeline solved the problem that highdimensional shape descriptors(HDSD)are difficult to use as classification features.At the same time,this classification pipeline well preserved the spatial distribution pattern of surface-based high-dimensional morphological features,and achieved better results than current state-of-the-art morphological classification methods,even it only used featured derived from a pair of subcortical structure.Finally,this classification pipeline is highly extensible in future research,so it can generate universal value in the diagnosis and classification of other brain diseases.