Research On Computer Aided Diagnosis For Alzheimer’s Disease Based On PCANet-BLS

Alzheimer’s Disease(AD)is a kind of dementia that occurs in the elderly frequently and is incurable currently.Only early detection and intervention can be used to delay the progression of the disease.Mild Cognitive Impairment(MCI)patients are between healthy elderly(Health Control,HC)and AD patients and have a high probability of transforming into AD patients.Therefore,the correct diagnosis of AD patients and MCI patients is particularly important.This work mainly uses Magnetic Resonance Imaging(MRI),and based on the prior knowledge that the structure of the hippocampus of AD and MCI patients has atrophy compared with healthy elderly people,the PCANet-BLS network model based on the principal component analysis network(PCANet)and Broad Learning System(BLS)was proposed to classify AD patients.The main work are as follows:1.269 AD cases,299 MCI cases,and 194 HC cases MRI data were collected from ADNI International Public Dataset and the First Affiliated Hospital of Guangxi Medical University and preprocessed.2.Constructing a PCANet-BLS model which uses unsupervised network PCANet to extract image features and then input these features into BLS for classification.Through a series of experiments based on the ADNI dataset,various parameters such as the number of the network layers of PCANet-BLS are determined,and then the ten-fold cross-validation method is used to test the performance of the model.The experimental results show that the classification accuracy rates of AD/HC,MCI/HC,AD/MCI groups were 97.08%,95.65%,91.60% respectively based the PCANet-BLS.3.This work also proposed to use an incremental learning algorithm of adaptive enhancement nodes to determine the number of enhancement nodes of BLS to ensure the classification performance of BLS.In addition,the influence of loss functions constructed by different regularization methods on BLS classification is also studied.The improved PCANet-BLS model was used to verify the clinical data and the accuracy of this method on the classification of AD/HC,MCI/HC,AD/MCI group reached 93.84%,90.26%,89.23% respectively,indicating the robustness and effectiveness of this method in AD classification tasks.In summary,this paper proposes a simple structure method for computeraided diagnosis of AD.The method has a short training period and does not require the size of the data set.The experimental results on the two datasets and the results compared with other methods also verify the feasibility and effectiveness of the method proposed in this paper.