In recent years,it was indicated that neuroimaging can be the potential tool to diagnosis Alzheimers disease (AD) and Mild cognitive impairment (MCI).To Accurately identify individuals at risk of developing disease,as AD,MCI shares cognitive and brain features with aging and neuroimaging is a kind of high dimension data which is impossible for traditional analysis.In this paper,we applied convolution neural network (CNN),a kind of deep learning method,for high dimension variables and shared characters from neuroimaging to execute three-way and binary disease classification tasks.We used the dataset on ADNI to perform the CNN model and compared the CNN with other traditional models.In this study,we collected the aging people over 55 years old whose brain is shrinking,and normal controls (NCs),MCI,AD were 159,157,153,respectively.In three-way classification of AD vs.MRI vs.with 3D-CNN,2D-CNN,Gaussian Process,we obtained the maximal accuracies of 85.5%,72.21% and 69.78%,respectively.This paper demonstrated high potential of using deep learning architectures opening new ways for medical diagnostic imaging especially from brain disorders.
Alzheimers Disease Mild Cognitive Impairment Deep learning Convolution Neural Network Magnetic Resonance Imaging
Wei Feng Nicholas Van Halm-Lutterodt Hao Tang Yuan Ma Zhiyuan Wu Ruoyao Cao Erlin Yao Xiuhua Guo
Department of Epidemiology and Health Statistics, School of Public Health,Capital Medical University Department of Neurosurgery, Beijing Tiantan Hospital, Capital Medical University, Beijing, China;Dep Institute of Computing Technology Chinese Academy of Sciences, Beijing,China Beijing Institute of Geriatrics, Beijing Hospital, National Center of Gerontology, Beijing, China Department of Orthopaedics and Neurosurgery, Keck Medical Center of USC,University of Southern Calif