Deep Learning Based Diagnosis Method For Alzheimer's Disease Using MR Images

Alzheimer's disease is one of the most common chronic diseases in the middle and old age,accounting for about 50%-60%of senile dementia.At present,there are about 50 million people suffering from Alzheimer's disease in the world.At present,there are about 40 million people suffering from Alzheimer's disease in the world.With the increase of the average life span of the global population,the aging trend is gradually obvious.As a result,it is estimated that the population of Alzheimer's disease patients has doubled in 20 years.At present,the main clinical treatment for Alzheimer‘s disease is to reduce the symptoms and delay the development of the disease through the combination of drug treatment and non-drug treatment.Therefore,the early diagnosis of Alzheimer's disease and its precursor symptoms is of great clinical significance for the treatment of patients.MRI diagnosis is one of the common diagnostic methods of central nervous system diseases.Through different scan sequences,MRI can not only provide morphological information of the brain,but also be used to evaluate the chemical composition status and functional status of different brain regions.Therefore,the role of MRI in the clinical diagnosis of Alzheimer's disease and its prodromal symptoms has been a hotspot in research.In this paper,we studied the influence of Alzheimer's disease on brain structure and the corresponding computer-aided automatic diagnosis method by using two imaging modality,i.e.,T1-weighted structure MRI and resting state functional MRI.Specifically,1)using the T1-weighted structural MR images of Alzheimer's disease,amnestic mild cognitive impairment and normal cognitive subjects collected by Shanghai Mental Health Center,we measured the cortical thickness and surface area of each brain area,and analyzed the brain regions of Chinese people with Alzheimer's disease and its precursor symptoms that were greatly affected by the disease.2)Using T1-weighted structural MR images in the public dataset ADNI,a new type of computer-aided diagnosis model based on coupled convolution and graph-convolution neural network is constructed,to process the image information of the lesion area and the spatial distribution of the lesion area.In this dataset,compared with conventional methods and state-of-the-art methods,our proposed method achieves the best performance on diagnosis tasks of Alzheimer's disease subjects/normal cognitive subjects,as well as progressive mild cognitive impairment subjects/stable mild cognitive impairment subjects.3)Using the resting state functional MR images in the ADNI dataset,we extract the dynamic functional connection matrix,and construct a novel graph-convolutional long short term memory network,which can be used for diagnosis tasks of Alzheimer's disease subjects/normal cognition subjects.In order to provide more effective information for the diagnosis of diseases,we introduce a hierarchical feature weighted sharing based multi-task mechanism,to explore the inherent relationship between diagnosis task and assistant tasks.At present,the highest diagnostic accuracy has been achieved in the ADNI dataset compared with other fMRI-based methods.Our works in this paper provide a valuable reference for the early diagnosis of Alzheimer's disease using MRI,and provides an algorithm basis for the implementation of high-speed and accurate computer diagnosis algorithm in the future.