Temporal And Spatial Analysis Of Brain Magnetic Resonance Images Based On Deep Neural Network

In the past decade,neuroimaging technology has made great progress.Because the techniques can fetch both structural and functional information of the brain tissue under non-invasive or minimally invasive conditions,Magnetic Resonance Imaging(MRI)and Computed Tomography(CT)have been widely used in the diagnosis of brain diseases.Doctors observe the structural and functional information of the brain tissues from the brain image data,and learned the changes of the patient’s lesions as the disease progressed according to the longitudinal images from multiple time points.Alzheimer’s Disease(AD)is one of the common brain diseases.At present,there is no effective means to cure AD.Early diagnosis of AD based on MRI brain image analysis is important for pretreatment of the disease.The computer-assisted diagnosis(CAD system with image processing and pattern recognition algorithms extracts detailed features from medical images and makes accurate diagnosis of brain disease.It provides more effective suggestions for doctors in the diagnosis and treatment of brain diseases.Deep learning is one of the most popular machine learning techniques in recent years.With the improvement of big data,hardware technique and the continuous innovation of theories,deep learning methods have been used in the analysis and processing of images,video,voice,text and other data forms.In the medical field,thanks to the wide use of the medical data,deep learning methods have been applied to the analysis of various medical data,including electronic medical records,diagnosis and treatment data,medical images,etc.Based such background,we propose two brain MRI image analysis algorithms based on deep learning method,for whole brain tissue and local hippocampus images.And we apply the proposed algorithms to the diagnosis of AD.The a lgorithms were tested on the Alzheimer’s Disease Neuroimaging Initiative(ADNI)database.And the results show that the proposed algorithms have good performances on AD diagnosis.The main innovations of this paper are:1.We propose a longitudinal brain MRI images analysis algorithm.A deep learning model is built for RAVENS map data.The spatial structure features and time variation characteristics of longitudinal data are extracted and integrated,leading to the improvement of AD diagnosis.The classification results on the ADNI database are 91.33% for AD vs.NC,and 71.71% for pMCI vs.sMCI.2.We propose a longitudinal hippocampus sMRI image analysis algorithm.The deep 3D-DenseNet models and traditional shape analysis method SPHARM-MAT are combined in the spatial feature extraction process of the hippocampus.Longitudinal analysis is performed to improve the performance of the overall algorithm.The experiment results on the ADNI database is 93.18% for AD vs.NC classification,75.62% for pMCI vs.sMCI,and 75.24% for MCI vs.NC.