Research And Implementation Of Alzheimer’s Disease Auxiliary Diagnosis Model Based On Personalized Federal Learning And Deep Forest

Alzheimer’s disease(AD)is a neurodegenerative disease that,once diagnosed,is irreversible and is often characterised clinically by memory impairment,visual and spatial deficits,as well as changes in personality behaviour and other cognitive deficits severe enough to interfere with daily activities and significantly reduce quality of life.As the course of the disease evolves over time,this has led to an increase in the severity of the disease in patients themselves and,in the context of the current ageing world population and the long course of the disease,AD has seriously affected the lives of patients and their families worldwide.In the medical field,the progression of Alzheimer’s disease is divided into three distinct stages,namely cognitive normal(Cognitive normality,CN),mild cognitive impairment(Mild Cognitive Impairment,MCI)and the definitive stage of Alzheimer’s disease(AD).The hippocampus is one of the first brain regions to be affected in AD,and therefore hippocampal volumes on magnetic resonance imaging(MRI)are often used to aid in the diagnosis of Alzheimer’s disease.In this paper,we use structural magnetic resonance imaging to segment 3D hippocampal images and calculate the hippocampal volume and propose the Fe De Fo personalized federal deep forest framework for Alzheimer’s disease classification and prediction in order to protect data privacy.The main research contents and work of this paper are as follows:1)To address the current problems of deep learning in image processing relying on large amounts of data,requiring huge computational resources,and lacking theoretical explanations,and for the purpose of protecting patient data privacy,this paper proposes a personalized federated deep forest framework for Alzheimer’s disease diagnosis.In order to effectively protect the data privacy of the client,we use a federated learning framework to collaboratively train a gradient-propelled decision tree(GBDT)model on each client’s local data.In addition,to address data discrepancies between clients,we introduce a deep forest model to further exploit local data beyond local interactions and fuse it with the federally trained GBDT to personalize the service for each client.Experiments show that the model achieves 91.6% and72.4% classification accuracy for AD and CN and MCI and CN respectively under the validation set.2)In order to better evaluate the performance of the proposed Fe De Fo classification model,different federated aggregation algorithms are deployed into the model in this paper,and the impact of data distribution issues across clients on the classification of the model is also considered.Therefore,experiments are conducted on multimodal data such as images and text related to Alzheimer’s disease to verify the impact of different aggregation algorithms on the classification effect of the model,and a series of experiments have proved the robustness of the Fe De Fo classification model proposed in this paper.3)To facilitate self-testing by patients,or assisted guided testing by patients’ families,according to the proposed model,this paper designs and implements an AD-aided diagnosis system based on personalized federated learning and deep forest.On the main page of the system,users can choose to submit s MRI images,while the back-end of the system automatically calculates hippocampal volume and invokes a classification model to predict the patient’s condition,and feeds the prediction results into the page.In addition,the system also uses the MOCA scale to aid diagnosis.