Research On Key Technologies For Deep Learning-based Assisted Diagnosis Of Alzheimer’s Disease

Alzheimer’s Disease(AD)is an irreversible neurodegenerative disorder that imposes a significant economic burden on patients.Therefore,researchers aspire to diagnose individuals in the preclinical stages of AD and provide appropriate medical interventions in order to effectively delay the onset of the disease.In recent years,researchers have treated AD diagnosis as a closed-set classification problem,where all categories are known,and the training and testing sets share the same classes.This approach has limited the application of models in actual clinical environments.Although various open-set recognition techniques have been proposed in other fields,their direct applicability to AD diagnosis is challenging for the following reasons: 1)AD is a neurodegenerative disease characterized by progressive symptoms,making it difficult to distinguish between different stages,particularly in relation to early-stage conditions.2)AD diagnosis involves multiple strategies,which makes it challenging to establish a unified modeling approach.Therefore,existing open-set recognition techniques are not directly applicable to AD diagnosis.While effective diagnosis of AD in open clinical environments would benefit patients,the irreversible nature of AD highlights the importance of early detection and medical intervention during the stage of Progressive Mild Cognitive Impairment(p MCI).Early and accurate diagnosis during the p MCI stage can help delay disease progression and provide greater assistance to patients.Functional magnetic resonance imaging(f MRI)is a neuroimaging technique widely used to investigate brain activity related to Alzheimer’s disease.However,acquiring f MRI data poses challenges,and the limited availability of data can lead to overfitting in classification models.Additionally,current p MCI diagnostic models lack interpretability,making it difficult for clinical practitioners to accept them readily.Based on the existing issues in current research,this paper aims to propose a diagnostic model for accurate diagnosis of AD and MCI.The proposed model is based on abnormal patterns and is designed to accurately diagnose AD in real-world open clinical environments.Additionally,a hybrid model based on brain networks is introduced to accurately diagnose the progression of Mild Cognitive Impairment(MCI).The main research contributions of this study are as follows:1.To address the challenges of large data volume and high complexity in routinely collected electronic medical records,this study establishes a unified data representation framework that enables more effective and comprehensive utilization of clinical data.2.Existing AD diagnostic models are primarily designed for closed-set environments,posing limitations in real-world clinical applications.This study incorporates the subjects’ abnormal patterns into the Open Max model by emulating the focus of clinical practitioners.This approach may encourage researchers to reevaluate the behavior of clinical doctors in real clinical settings.Moreover,Open APMax can be seamlessly integrated into other models without requiring changes to the model’s architecture,demonstrating excellent scalability.Compared to existing open-set recognition algorithms,Open APMax exhibits superior performance in complex diagnostic tasks.3.Addressing the issue of limited sample size and potential overfitting in functional magnetic resonance imaging(f MRI),this paper proposes a novel hybrid model based on brain graph representation for diagnosing Mild Cognitive Impairment(MCI).The model utilizes a multi-head graph attention network(Graph Attention Networks,GAT)to learn the relationships between local regions of interest(ROIs)and employs a multi-layer perceptron(MLP)to capture global features,thereby supplementing the learning of classification features and reducing model overfitting.Experimental results demonstrate the superior performance of this hybrid model.Additionally,the study analyzes the network attention coefficients to identify and visualize the significant ROIs influencing MCI.