Voxel-Based Morphometric Study Of The Cerebellum Volume Changes In The Early Alzheimer's Disease

Alzheimer's disease(AD)is the most common neurodegenerative diseases in the elderly population.The clinical diagnostic criteria for AD rely mainly on neuropsychological behavior tests and clinical laboratory tests,but it is difficult to detect its pathological status.So far,whether it is the existing treatment of AD or the drug being developed,all are expected to delay the progression of AD in the prophase and early stage of AD.Therefore,it is extremely important to find a method that can accurately diagnose AD in the early stage,especially in the prodromal stage.Conventional magnetic resonance imaging(MRI)is widely used for brain morphologic evaluation.With the assistance of OASIS dataset,we have determined the boundary of cerebellum,and it is treated as the standard of segmentation.ITK-SNAP is used as interactive segmentation tools for cerebellum segmentation.Through these process,cerebellum label dataset is established,which has 81 samples.This dataset is used to evaluate automatic cerebellum segmentation method below and study the morphology of AD patients' cerebellum.Besides,considerin the dataset is not big enough,data augmentation methods are applied to extend the dataset,in order to make training and evaluation of deep learning based automatic cerebellum segmentation.Due to the time cost of interactive is too long,it is hard to fit the needs of large dataset analysis.Therefore,deep learning based automatic cerebellum segmentation method in MRI is proposed.Fully convolutional neural network is developed.Besides,through a lot of contrast experiment,light and shade change is found to be the best augmentation method for our proposed deep neural network.The proposed method get 0.95 of Dice coefficient and 0.9 of Io U on automatic segmentation,which significantly improve the efficiency of cerebellum segmentation.Voxel-based morphometry(VBM)is one of the most popular technique to study the segmentation of cerebellum.With the VBM method,we designed the data process procedure.Compared with the normal samples' cerebellum,it is found that AD group shows atrophy on left posterior lobe and right posterior lobe of cerebellum,which mainly focus on vermis and tonsilla.And patients in early AD are found that atrophy happens on right posterior lobe of cerebellum,which mainly focus on vermis.Through interactive segmentation,cerebellum label dataset is eastablised in this article.Besides,fully convolutional neural network based automatic cerebellum segmentation method is constructed.Finally,through VBM-based cerebellum morphometry analysis,we found that the volume of right posterior lobe of cerebellum and vermis is related to AD.The discovery above provides reference for clinical image diagnosis of early AD.