MRI Multi-modal Features-based Research For Affective Disorders Recognition

Magnetic resonance imaging(MRI)is a conventional tool for clinical diagnosis,and it is also the medical imaging data commonly used in brain science studies.Structural MRI contains a large amount of structural information about the target region,while functional MRI can capture the functional information of brain activation.These two kinds of information were extracted and analyzed quantitatively,and can be used to assist clinical diagnosis and make treatment strategies.In recent years,artificial intelligence technology has shown great application potential in intelligent industrial upgrading in many fields.The intelligent analysis of medical images combined with artificial intelligence technology has attracted the attention of many researchers and clinical staff."Radiomics" is a typical intelligent analysis method for medical images,and its effectiveness has been confirmed in the field of oncology.But,there are still few studies on the applications of radiomics in psychiatry.Following the rapid development of China’s economy,the prevalence of various mental disorders increased.The affective disorder is a typical mental disorder with a high incidence rate.However,at present,the pathogenesis of the affective disorder is not clear,and the diagnostic accuracy is not high enough.There are still not any accurate and objective imaging markers for diagnosis,causing missed diagnoses and misdiagnosis in clinical.Therefore,focusing on the diagnosis of affective disorder,this dissertation extracted quantitative information from MRI data,applied functional or structural features,constructed a medical intelligent recognition algorithm,and explored the pathogenesis of the affective disorder.The main contents and contributions of this dissertation are as follows:Firstly,this dissertation explored the feasibility of brain functional characteristics combined with radiomics analysis in the recognition of affective disorders.The functional MRI data of90 unmedicated patients with bipolar disorder(BD)and 117 healthy controls were collected.Four types of functional features were extracted.To discriminate the BD patients from healthy controls,the imaging classification model was constructed through several steps,including statistical analysis,embedded feature selection,and machine learning.Finally,this study achieved a classification accuracy of 80.5% in the test set(AUC = 0.838),which proved that the radiomics analysis can be used to identify the patients with affective disorder,and provided an effective modeling method for constructing and improving the recognition model of affective disorder by using functional features.Secondly,this dissertation explored the effectiveness of shape and texture features based on cortical gray matter structure for identifying the affective disorder.The structural MRI data of 114 patients with BD and 139 patients with major depression disorder(MDD)were collected.The standard anatomical AAL template was transformed into individual space through spatial inverse transformation.The individual brain was automatically divided into116 brain regions,then the shape,first-order statistics,and high-order texture features of each brain region were extracted.This study constructed a classification model using the shape,texture features,and GMV features of the core brain regions,and compared it with the model constructed only by GMV features.The accuracy of the final classification model in discriminating the subtypes of affective disorder reached 85.3%,and the performance was improved compared with the GMV feature model.This study showed that the shape and texture features of the cortex could be used in affective disorders.Subsequently,the structural MRI data of 80 patients with hippocampal sclerosis epilepsy and 80 healthy controls were collected to validate the research conclusion in epilepsy.Using the same method,the results verified the effectiveness of the shape and texture features of the cortex in exploring the abnormalities of the nervous system.Thirdly,this dissertation explored the multimodal and multi-type feature fusion strategy for affective disorder recognition.Resting functional MRI and structural MRI data from 136 MDD patients and 171 healthy controls were collected.Three types of functional features were extracted in functional MRI and GMV features were extracted in structural MRI.In these four types of features,feature selection was carried out respectively to choose the important features under each type,and then these four types of selected features were constructed into different feature combinations according to the feature types.Under each combination,a classification model to distinguish MDD patients and healthy controls were built respectively.The model based on multi-modal and multi-type features finally achieved high classification performance,and the AUC reached 0.916.This study finally found an optimal multi-modal and multi-type feature fusion strategy.Finally,this dissertation explored the influence of the brain segmentation template.In this study,three different types of segmentation templates were used to repeat the classification study of MDD patients and healthy people,and the same method was used under each type of template.It is found that the selected brain regions and the model performances under various templates were different.According to the results,it is speculated that the templates of different modals might be suitable for different types of analysis,which provides a reference basis for choosing the appropriate template for brain imaging analysis.In addition,the consistent analysis of different brain regions selected under different templates,and the application of different templates for analysis may reduce the differences among various studies.This dissertation fully analyzed the application methods of multi-type features extracted from different modal MRI in the identification of affective disorders,provided ideas for the construction of objective and accurate imaging markers of affective disorders,assisted in the application of clinical diagnosis,revealed the potential pathogenesis of affective disorders,and promoted the development of precision medicine.In addition,the discussion of different templates provides reference views for eliminating the differences among various studies.