Early Diagnosis Of Alzheimer’s Disease Based On Slice Attention Model

Alzheimer’s disease(AD)is a progressive,incurable neurodegenerative disease.Early detection can prevent further brain cell damage in patients,which is consistent with the concept of "active health".Currently,there is still a lack of methods to diagnose Alzheimer’s disease at an early stage,which leads to patients being diagnosed at a relatively late stage,making treatment difficult and leading to a lack of milestones for drug development and missing the optimal time point for drug treatment.Therefore,there is an urgent need for a rapid,fully automated assessment method to accelerate the diagnosis of Alzheimer’s disease.Unlike previous approaches based on structural magnetic resonance imaging(s MRI),which simply transfer the neural network model that is better in natural image classification to medical classification,the method proposed in this paper fully considers the characteristics of structural magnetic resonance imaging(s MRI)and proposes a deeply residual network model for 3D abnormality perception based on 3D ResNet,including 3D residual squeeze and excitation module(RSE),and a recurrent slice attention module(RSA).Inspired by the effectiveness of squeeze and excitation modules in natural image classification,RSE aims to integrate the squeeze and excitation modules into the residual blocks to capture the importance of different channels,enhance important feature information,suppress useless information,and filter out the most informative features.RSA consists of a slice attention module(SA)in coronal,sagittal and axial directions,which aims to model 3D MRI images as slice sequences to capture the long-distance dependence of different slices in different directions.RSA combines the context information of the abnormal area with the local information and spatial information to obtain better classification performance.Experimental results show that our proposed method outperforms state-of-the-art models in classifying normal controls(NC),early mild cognitive impairment(EMCI),late mild cognitive impairment(LMCI)and Alzheimer’s disease(AD)on the ADNI dataset,achieving87.5% accuracy.The results of Grad-CAM visualization also show that our method can successfully highlight brain regions that contribute more to classification performance.