Research On Alzheimer’s Disease Diagnosis Based On Multiple Information

Alzheimer’s disease is a neurological disorder that leads to a gradual decline in memory,thinking,behavior,and emotions.Although there is currently no complete cure,early diagnosis can help doctors and patients take treatment measures to slow down the condition’s deterioration.The traditional diagnosis of Alzheimer’s disease requires multiple examinations,including brain scans,blood tests,and cognitive behavioral assessments.However,integrating these data requires doctors to have significant clinical experience.With the development of deep learning technology,automatic diagnosis methods for Alzheimer’s disease based on large amounts of data have become a research hotspot.This article focuses on the application of deep learning models in the diagnosis of Alzheimer’s disease,based on the analysis of a large amount of clinical data and brain imaging data.(1)This study proposes a classification model MM-Net that can extract features from multiple paths and scales,aiming to address many models that fail to fully utilize the anisotropy of different axial data of MRI images.The same MRI image is input into the model from three directions,namely,transverse,coronal,and sagittal planes,and the data is obtained through different branches to obtain multi-path features and fused to obtain more comprehensive information.Additionally,the study adopts a multi-scale feature extraction strategy,introduces attention mechanism CBAM,and an adaptive feature fusion mechanism to preserve more local details and focus the model on key areas.Experimental verification using the public dataset ADNI shows that the AUC value reaches 0.914,which is 0.03 higher than the basic 3DCNN model.(2)This study proposes a solution to the problem of low model efficiency caused by using an entire image as input,by diagnosing using several important patches instead of the entire image.Firstly,the study extracts the ROI(region of interest)of MRI image data and calculates the gray matter volume of the ROI.Then,the two sets of gray matter volume data of ROI are statistically analyzed using a dual sample t-test,and the resulting t-value graph shows the difference distribution between the two sets of data,sMCI and pMCI.Then,a patch selection strategy based on t-value graphs is designed to filter several important patches without overlapping too many regions.Finally,the study constructs an integrated learning classification model t-INet based on CBAM-CNN.The experimental results show that,without significantly reducing the classification accuracy,the amount of model parameters and operational efficiency have been greatly improved.Compared to the CBAM-CNN model using the entire image as input,t-INet has an AUC value increase of 0.009 and a parameter amount decrease of 36.7%.(3)This study designed and implemented a diagnostic system for Alzheimer’s disease based on theoretical research,using the Qt framework,including three modules:data management,model management,and auxiliary diagnosis.The system has a good humancomputer interaction page,and the data management module can generate a subset of data for in-depth learning,and perform data preprocessing operations.The model management module can create new models for training and testing,and save experimental results under various parameters.The auxiliary diagnosis module can provide auxiliary diagnosis decisions based on the patient’s MRI images and clinical data.