Research And Implementation Of MRI Image Classification In Alzheimer’s Disease Based On Deep Learning

Alzheimer’s Disease(AD)is an irreversible neurodegenerative disease commonly seen in the elderly.The main characteristics are progressive memory loss and cognitive impairment,accompanied by personality changes and other characteristics.Mild Cognitive Impairment(MCI)is an early pathological condition between AD and Normal Control(NC).There is no effective drug that can completely cure AD patients.The pathological conditions of patients can be better controlled If prophylactic intervention can be carried out.Therefore,early and accurate diagnosis of MCI and AD patients is of great practical significance to slow down the development of the disease.Magnetic Resonance Imaging(MRI)is a non-invasive imaging technique,which has outstanding imaging effects on the changes of brain structure in AD patients.Deep learning can avoid complex artificial selection of features and high-dimensional disasters of data in the medical classification task.Therefore,the methods based on deep learning to classify MRI images of Alzheimer’s disease is proposed:1.Classification of Alzheimer’s Disease based on block confidence and densely connect network.Firstly,the original MRI data are preprocessed,and the slices were extracted along the sagittal plane,coronal plane and transverse section.According to the distribution of the lesion area in slices,three 2D slices were selected for each 3D MRI sample that could clearly identify the lesion features.Secondly,a densely connect network model based on 2D data is designed by introducing dense connection.The selected 2D MRI slices are automatically cropped to augment the dataset through the block confidence algorithm.Finally,the segmented sub-images are trained for the model.The experiment results show that the proposed data augmentation algorithm solves the problem of image detail loss caused by traditional augmentation methods,and achieves 90.16% and 78.03% accuracy in AD vs.NC and MCI vs.NC classification tasks respectively.2.A classification algorithm for Alzheimer’s disease based on attention mechanism and residual-dense network.Considering the mechanism that the original low-level information can be used several times by the residual network,a classification method based on the residual-dense network is designed combining the dense connection and residual fusion.Secondly,considering that different layers of the network can be assigned different weights by the attention mechanism to focus on the image feature regions which has high correlation with the classification task,an attention mechanism applied to the multi-layer connected network is proposed based on the residual-dense network.Finally,the residue-dense attention network embedded with attention module is designed to classify Alzheimer’s disease.Finally,the accuracy of AD vs.NC and MCI vs.NC were 94.71% and 81.90%,respectively.In summary,several convolutional neural network models with different architectures based on structural MRI images are designed and implemented to Alzheimer’s disease classification.The effects of different network design ideas and data augmentation methods on the classification results are analyzed in the experiments.The experimental results show that the proposed classification model of Alzheimer’s disease based on block confidence augmentation algorithm and residual dense attention network yields the best performance,which solves the problem of insufficient use of feature information in convolutional neural network,and can obtain more effective feature image information,the accuracy of AD vs.NC and MCI vs.NC classification tasks are improved.