Classification Of Alzheimer’s Disease Based On Magnetic Resonance Images Of The Human Brain

As the world’s population ages,the importance of diagnosis and effective intervention for Alzheimer’s disease is becoming increasingly important.However,the current diagnosis of Alzheimer’s disease still relies on clinical symptoms and the subjective judgment of specialized physicians,which has great limitations.Therefore,the objective diagnosis of Alzheimer’s disease has attracted widespread attention from academic and medical communities,and in line with this,the classification of Alzheimer’s disease based on magnetic resonance images of the human brain has become one of the research hotspots in the field of brain science in recent years.Currently,a large number of structural MRI-based Alzheimer’s classification studies have been conducted internationally,but the strategies used in these studies have some shortcomings,such as not effectively utilizing the topological properties of human brain networks,high model complexity,and difficulty in capturing association information between two-dimensional image slices.In addition,exploring the physiological mechanisms that advance Alzheimer’s classification is also a current research concern.In addition,investigating the physiological mechanisms that advance Alzheimer’s classification is also a focus of current research.To address the above issues,three works were carried out in this study as follows:(1)Alzheimer’s disease classification based on attention-guided hybrid convolution-graph convolution network.This work uses 3D magnetic resonance images as input.First,we extract the brain structural features contained in the human brain MRI images using a convolutional neural network and generate an attentional map.Then,we identify the key brain regions for classification based on the guidance of the attentional map and construct a graph convolutional network to extract the topological structural features of the human brain.Finally,we use feature fusion to achieve the complementarity of whole-brain image features and topological structure features to improve the classification accuracy.The experimental results show that combining the attention mechanism with the graph convolutional network can better achieve the classification of Alzheimer’s disease,and the patient’s brain lesion regions can be effectively localized based on the attention map.(2)Alzheimer’s disease classification based on multi-view dynamic images and clinical features.In this work,we first synthesize dynamic images of 2D MRI slices in coronal,sagittal,and axial views,and used pre-trained models to obtain dynamic image features for each of the three views.Then,the clinical features were fused with the image features in a channel attention manner for the classification of Alzheimer’s disease.Finally,the prediction scores of multiple viewpoint models were integrated to obtain the final classification results.The experimental results show that the classification accuracy of Alzheimer’s disease can be effectively improved by the strategies of dynamic images,multiple views,and fusion of clinical features.(3)Alzheimer’s disease classification based on structural connectivity of grey matter density histograms.This work draws on the idea of functional connectivity used in functional magnetic resonance image analysis.In this study,we identify multiple regions of interest based on templates and extract the gray matter density histograms of the relevant regions,and then four histogram structural connectivity features can be obtained based on the correlation between the histograms.Finally,we classify Alzheimer’s disease based on these features using a classical classification algorithm.The experimental results show that the novel structural connectivity features proposed in the study combined with machine learning methods can obtain good results for Alzheimer’s disease classification,and the classification model constructed in this study facilitates the exploration of potential physiological mechanisms that advance Alzheimer’s disease classification.