Classification Of Brain Diseases Based On Multimodal Imaging And Machine Learning

In recent years,with the rapid development of economy and society and the acceleration of the pace of life,the prevalence rate of mental illness has been increasing year by year.According to the latest data from the Ministry of Health,approximately 17% of adults in my country suffered from mental illness,and the burden of mental illness ranks first in the country’s total disease burden,which has surpassed cardiovascular,cerebrovascular,respiratory and malignant diseases.However,the current diagnosis of mental illness mainly relies on experienced clinicians and the evaluation of clinical scale,and there is a lack of objective diagnostic criteria.At the same time,it is very likely to cause misdiagnosis and missed diagnosis due to the overlap of clinical symptoms and the subjectivity of the doctor’s diagnosis.Therefore,there is an urgent need to realize a computer-aided diagnosis system for accurately diagnosing mental diseases.With the continuous development of medical imaging technology,the exploration and auxiliary diagnosis of the pathogenesis of mental diseases based on neuroimaging data has become a research hotspot in recent years.Many studies have used single modal imaging data to diagnose mental illness,but the results have been unsatisfactory.Multi-modal neuroimaging data can make full use of the information of each mode to complement,so as to obtain more comprehensive disease information.Therefore,this thesis focuses on the repeatability of features,the instability of classification performance,and the use of single-modal image data to diagnose mental illness but cannot fully understand the comprehensive information of the diseased area.We use graph theory and machine learning algorithms to slolve the above problems.The main research results include the following two aspects:a)In view of the repeatability of features and the instability of classification performance,this paper proposes a data-driven feature stability selection method to ensure the repeatability of features and the stability of classification performance.The main idea of the data-driven feature stability selection method is as follows: In a given series of feature stability selection methods,graph theory is used to evaluate the feature repeatability,average accuracy and feature stability of various feature selection methods.To explore the relationship between the stability of different feature selection methods is to select the best feature stability method.Experimental results show that the data-driven feature stability selection method proposed in this paper can select the optimal feature stability method,and at the same time select the feature with the highest repeatability and the most stable classification performance.b)For schizophrenia and bipolar disorder,due to the overlap of clinical symptoms and the loss of some cognitive functions,it is difficult to use single modal imaging data for the diagnosis of mental disorders.At the same time,comprehensive information on the lesion area is not fully understood,resulting in the failure to achieve accurate diagnosis.This paper presents a multi-modal and machine learning approach for the classification of schizophrenia and bipolar disorder.Specifically,we used multimodal data including structural magnetic resonance imaging data and resting functional magnetic resonance imaging data to categorize schizophrenia and bipolar disorder using support vector machines,logistic regression,random forest,and deep neural networks.The experimental results show that the performance of multi-modal data based on the same classifier is better than that of single-modal data.According to the correlation between the prediction probability and the clinical scale,the reliability of the classification results is reflected laterally.