Alzheimer’s Disease Image Classification Algorithm Based On C3D-BiLSTM Network And Cost-sensitive Learning

Alzheimer’s disease(AD)is an irreversible neurodegenerative disease of the brain.As the disease worsens,patients usually suffer from cognitive decline,memory decline and other symptoms.In severe cases,they may lose the ability to take care of themselves,which greatly increase the burden of life on patients and their families.As the problem of population aging in our country becomes more and more serious,the incidence of AD is gradually increasing.Disease intervention for early AD patients can effectively improve the survival cycle and quality of life of patients.Therefore,accurate diagnosis of AD at different development stages is particularly important.In recent years,the development of medical imaging technologies such as CT,MRI and fMRI have also provided favorable conditions for the research of disease diagnosis.Using these data for computer-aided diagnosis can not only reduce the burden on doctors,but also reduce the occurrence of misdiagnosis and missed diagnosis,which is of great significance for the development of the patients and the whole society.This thesis studies AD classification using functional magnetic resonance imaging(fMRI)data,which is the 3D sequence having spatial and temporal characteristics,also the problem of category imbalance.The main work includes:(1)Aiming at the feature of fMRI data,the C3D-BiLSTM network combined with the attention mechanism is used to classify AD data.This model can extract spatial and temporal features from raw fMRI data that are not sliced.Since the key segments of fMRI reflecting the lesion information are at the different positions of the sequence,a self-attention mechanism is introduced to pay more attention to the learning of important sequences.Experimental results show that this model shows a better classification performance than the model with separated data slicing and feature extraction.(2)Aiming at the imbalance characteristics of fMRI data,cost-sensitive method is introduced to learn collaboratively with the network.The outputs of the last layer are weighted to increase the penalty for misclassification of minor categories,and to improve the classification accuracy of minor categories,so as to ensure the accuracy of the classification of the major categories.Experimental results show that the algorithm alleviates the problem of the category imbalance of AD data on classification,and reduces the misclassification rate of minor categories in AD data.