Magnetic Resonance Imaging Based Methods For Auxiliary Diagnosis Of Alzheimer’s Disease

Alzheimer’s disease(AD)is a mentally degenerative disease that causes great harm on patients’ memory,judgment and language skills.Clinically,diagnosing Alzheimer’s disease mainly relies on the doctor’s ini-tial auxiliary diagnosis combining neurology and medical imaging technol-ogy,followed by persistent tracing analysis which may cause huge pressure on human and medical resources.Experiments show that hippocampus means a lot to auxiliary diagnosis of Alzheimer’s disease as the first af-fected region during the disease.Achiving accurate hippocampus segmen-tation and combining the segmentation of hippocampus for disease clas-sification will better help auxiliary diagnosis of Alzheimer’s disease.Re-cently,with the continuous development of neuroimaging technology and deep learning,it provides a data basis and technical means for the diag-nosis of Alzheimer’s disease.Therefore,based on the structural magnetic resonance imaging data of Alzheimer’s disease,this study develops an aux-iliary diagnosis method for Alzheimer’s disease.The main research work is as follows:(1)Accurate hippocampus segmentation provides an important basis for clinical diagnosis and greatly relieves the medical pressure.At present,most of the research based on hippocampus segmentation only considers the overall distribution of the hippocampus without considering the bound-ary,structure and other multi-level details,and the utilization of multi-scale features is not sufficient.Therefore,in this study,we propose PMF-FL Net.First of all,we adopt the parallel primary and auxiliary net with shared en-coder so that the parallel net focuses on the corresponding area for jointly optimizing the encoder information.The primary net focuses on the region and boundary information,and the auxiliary net focuses on the structure in-formation.And we introduce the region,boundary and structure loss form-ing a joint loss to ensure the effectiveness of the segmentation.Finally,the multi-scale feature learning module is improved to optimize the interaction of multi-scale information to supplement the lack of information in the de-coding layers of each layer.The experiments are evaluated on the Har P dataset,and the results show that the proposed method is not only better than the four popular deep learning-based segmentation methods,but also has obvious advantages compared with the popular tools.(2)Accurate prediction of Alzheimer’s disease enables patients to real-ize the disease as soon as possible and thus receive timely treatment,which is the most important part of auxiliary diagnosis.However,most of the existing classification models are based on the whole structural magnetic resonance imaging,without incorporating important regions such as hip-pocampus for analysis guidance.Therefore,this study proposes DABHS Net.This method constructs a hippocampus segmentation-aided net and a classification net.The classification net extracts the high-level seman-tic features in the segmentation-aided net and analyzes them to obtain the classification results.The classification loss and the segmentation loss are combined into a joint loss for optimizing network,and the segmentation task is used to assist obtaining accurate classification as prediction results.At the same time,a global-pooling attention mechanism is introduced to effectively guide the features extracted in the classification net combined with the hippocampus.The method is experimentally evaluated on the Har P dataset and ADNI-1 dataset.The experimental results show that the method not only performs better in segmentation results,but also outper-forms the current classical methods in the Alzheimer’s disease classification task with robustness and adaptability,and it can be transferred between dif-ferent datasets to effectively assist the diagnosis.At the same time,for the dataset without manual segmentation of the hippocampus,the PMS-FLNet can be used to segment the dataset by the hippocampus label,and then combined with DABHS Net to obtain a more effective Alzheimer’s dis-ease classification,which provides a technical support for further research on auxiliary diagnosis of Alzheimer’s disease.