Brain Disorder Assisted Diagnosis Based On Functional Magnetic Resonance Imaging And Machine Learning

The brain is an important organ in the human body.However,the existence of various brain diseases threatens people’s health and brings serious burden to families and society.Functional magnetic resonance imaging(f MRI)is a common brain imaging techniques used in the field of medical.This technique was widely used in the field of cognitive neuroscience because of its non-invasive,simple experimental procedure and easy practical application.Meanwhile,machine learning is favored by many researchers for its powerful learning ability.In particular,deep learning could achieve automatic extraction and analysis of disease features and it has also been widely used in medical image processing for aiding disease diagnosis in recent years.Currently,the idea of using machine learning to assist in the diagnosis of brain diseases has been widely adopted.However,the results of previous studies are still characterized by insufficient precision,poor diagnosis of diseases,and obscure disease biomarkers.In this thesis,two typical brain disorders,Alzheimer’s disease and autism spectrum disorder were investigated using f MRI and combined with machine learning.We enhance the diagnostic performance of two types of diseases through effective extraction of important features and mining of new diagnostic markers.The following two main tasks were included:(1)An Alzheimer’s disease diagnosis method based on multiple feature selection and machine learning was proposed.Firstly,the correlation between different regions of the brain was calculated to characterize their functional connectivity strength,and the functional connectivity situation was used as a basis for disease diagnosis and treatment.Subsequently,a threshold definition method was used for the removal of weak connections,an element reduction method for the removal of duplicate feature elements,and the LASSO selection for the multiple screening of important features,retaining those that are more valuable for disease diagnosis performance.Finally,two machine learning methods,support vector machine and extreme gradient boosting algorithm were used for disease classification and prediction.The method was applied to the diagnosis of Alzheimer’s disease and obtained good classification performance.The results demonstrated the effectiveness of the feature selection method for disease prediction.(2)A disease diagnosis method based on multivariate dynamic features and deep learning was proposed.The method exploited the unique temporal characteristics of functional magnetic resonance images for multivariate dynamic feature mining,while diagnosing Alzheimer’s disease and autism spectrum disorder brain disorders with the help of deep learning.Firstly,spatially independent component analysis was performed to obtain some resting state networks,followed by screening of important components and removal of noisy signals.Then the sliding window was used to temporally segment the time series of each component to obtain the sub-time series,and the correlation of the sub-time series within each window was calculated as a dynamic functional connectivity representation among the networks.Finally,a long and short-term memory network sensitive to temporal signals was used for classification.The experimental results show that the method proposed in this thesis could make effective prediction of diseases and outperforms existing algorithms.In addition,during the experiment,it was found that Alzheimer’s disease patients have abnormalities in the dynamic connection process between the hippocampus region and other networks,while autism spectrum disorder patients exhibit abnormalities in the attention network.This thesis investigates two methods based on the combination of f MRI and machine learning for brain disease diagnosis,and applies the two methods to the diagnosis of two typical brain diseases,obtaining satisfactory diagnostic accuracy.This study broadens the avenues of medical images for disease diagnosis,which is important for the diagnosis and treatment of brain diseases such as Alzheimer’s disease and autism spectrum disorders and the discovery of effective biomarkers for the diseases,and also contributes to the development of computer-aided diagnosis technology.