Research On Mental Diseases Prediction Methods Based On Functional Magnetic Resonance Imaging

Mental diseases,including autism spectrum disorder,Alzheimer’s disease,and major depression,have been on the rise in recent years,seriously affecting tens of millions of families.At the same time,functional magnetic resonance imaging(f MRI),as noninvasive imaging technology,can record the whole brain activity and the response of patients and has become a key research method in contemporary neuroscience to study the cerebral cortex network.At present,the development of machine learning algorithms provides unprecedented opportunities for functional magnetic resonance imaging data analysis.How to assist doctors in the early screening of mental diseases through big data analysis and modeling has important practical significance and urgent needs.Therefore,in this thesis,the functional magnetic resonance imaging(f MRI)data of mental diseases available in-house and public were used to construct an initial screening diagnosis model based on image processing,multiple kernel learning,spatial mapping,and other methods.Firstly,a multi-sequence and multi-scale asynchronous correlation statistical analysis method was proposed to analyze the statistical characteristics of brain regions in functional magnetic resonance imaging of psychiatric diseases.Unlike previous methods,The model innovatively introduces temporal sequence information,context information,and asynchronous correlation in functional magnetic resonance imaging data,and excavates the correlation information among various brain functional regions from multiple angles.In this study,the probability distribution information of co-expression between different brain regions was counted,and the frequency information of co-expression between any two brain regions in the sample was calculated under different time delays.Secondly,a new framework for screening and diagnosis of mental diseases based on multiple kernel learning is proposed.In this framework,a time series kernel function based on the probability distribution of state transition in one brain region and the probability distribution of cooperative expression in multiple brain regions is constructed,and the Pearson correlation coefficient and dynamic time warping method are combined.A high-precision screening and prediction model of mental diseases was established by multiple kernel learning weighted fusion.The model describes the changes in functional connections between different brain regions and the abnormalities of single brain regions in the form of probability statistics mode,which breaks through the traditional method that only one or more scalars can be used to define functional connections.It can more accurately locate and distinguish different brain regions and abnormal patterns based on large samples.Finally,a new multi-instance alignment space mapping method based on random neighborhood embedding is proposed.This method can map multiple isomorphic Spaces to the same target space according to multi-instance alignment constraints.The visualizations of all samples at the brain region level were obtained.A high-precision ensemble classifier is constructed by extracting the subspace of the homonymous brain region and adopting an ensemble learning strategy.The model was tested on public data sets and achieved good disease prediction accuracy.In conclusion,this study uses machine learning computer technology and biomedical knowledge to work on f MRI image data,providing ideas for early screening and accurate prediction of mental diseases.