Research And Application Of Brain Multi-Modal MRI Processing Method

Multi-modal magnetic resonance imaging(MRI)provides us with imaging information from the structure to the function of the brain.Many studies have shown that some mental or neurological diseases are related to changes in the brain structure and function.These changes may be intrinsically related to each other.Analyzing the multi-modal magnetic resonance imaging provides the possibility to reveal the relationships between structure changes and brain function changes in some diseases.In addition,the comprehensive medical imag information provided by the multi-modal magnetic resonance imaging plays an important role in exploring the working of brain.Multi-modal magnetic resonance imaging has the advantages of information complementarity,but how to use the data with different attributes to comprehensively reflect the biological information of the subjects is still an urgent need and a major challenge.On the other hand,among numerous serious diseases that plague human beings,obesity has being one of the most concerned due to its epidemic and easy to cause other intractable diseases.Researches hope to explore the abnormal brain function or structure in subjects with obesity to reveal the neural mechanism of obesity,however,few researches applied joint analysis from brain structure to function.Therefore,this paper focuses on the research of the multi-modal magnetic resonance imaging processing methods and applications in the following four aspects.(1)By analyzing the T1-weighted image and resting-state functional magnetic resonance image(fMRI)separately,the changes of brain structure and function in subjects with obesity and the effect of bariatric surgery on these changes were investigated.It was found that subjects with obesity had decreased resting-state activity in the orbitofrontal cortex and decreased gray matter volume in the cerebellum and occipital-parietal lobes.Furthermore,these changes were recovered following the weight loss after the surgery.Our results reveals the plasticity of brain structure and brain function in subjects with obesity,which is significant for revealing the mechanism and treatment of obesity.(2)Parallel independent component analysis(parallel-ICA)was used to analyze the associations between the changes of brain structure and changes of brain function in subjects with obesity.It was found that the changes in brain function were significant associated with changes in brain structure.The results for the first time reveal that there is strong coupling between functional changes and structural changes of brain in subjects with obesity,though the plasticity of brain structure or brain function is not supported by the results.The couplong reveals the inner link between alterations of brain structure and brain function in subjects with obesity,playing an important role in understanding the neural mechanisms of obesity.(3)To solve the defects in the current multi-modal magnetic resonance imaging analysis methods,we proposed a new method for multimodal joint analysis.The method was included two main parts.First,non-negative blind source separation was applied to the image of each mode as the measured MRI signal is comprehensive reflection of various physiological information and noise.Based on the fact that little correlations between the volume of different brain regions,non-negative matrix factorization was applied to the T1-weighted image to decompose the image into the linear superposition of several non-negative basic images.For the first time,the improved non-negative independent component analysis was applied to the fMRI data to decompose the image into the linear superposition of several non-negative independent images as significant correlations between different brain regions.The two non-negative blind source separation methods was applied to two different MRI modes to separate the mixed images into series physiologically explainable components(non-negative components).Simulation data was used to verify the feasibility of these two blind source separation methods on the strucutural image and functional image respecitively.The second major part of the methodology is the feature selection of the decomposed basic images.Considering the various modal properties and large amounts of basic images,we applied the the integrated machine learning method—random forests to select the feature images from the decomposed basic images as the random forests method is suitable for processing large amounts of features and has no requirements to the properties of the features.To obtain the relationships between different image modals,the correlation coefficients between the selected features were calculated.(4)The established multi-modal images conjoint analysis method was applied to the fMRI and T1-weighted images to investigate the the neural changes in subjecects with obesity.It was found that there was coupling between altered brain volume and disordered resting-state brain activity.In addition,the spatial distribution of the selected components from fMRI and T1-weighted images has a certain overlap.The results obtained by our established analysis method verified the coupling of brain structural changes and brain functional changes in subjects with obesity,as well as the plasticity of such structural and functional changes,which indicates that the multi-modal MRI processing method we established has higher detection efficiency than the existing methods.