Early Diagnosis Of Alzheimer’s Disease Based On Transfer Learning And Visualization Aided

At present,more and more elderly people are suffering from Alzheimer’s Disease(AD).Clinical studies have shown that the probability of mild cognitive impairment(MCI)conversion to AD is high.However,but it can be cured if it can be treated at the MCI stage.Therefore,timely detection of symptoms is essential for effective treatment and prevention of brain tissue damage.The task of this thesis is to improve the accuracy of t diagnosis based on magnetic resonance imaging(MRI)medical images greatly.Among various identified biomarkers,magnetic resonance imaging(MRI)is widely used to predict AD or MCI.While good characterization is critical to classification performance,almost all previous studies often rely on human markers to find regions of interest(ROI)that may be associated with AD,such as the hippocampus,amygdala,and precursors.This program requires domain knowledge and is both expensive and tedious.In this thesis,the data augmentation method is used to expand the size of the Original data set.We train data set of the MRI images from the Alzheimer’s Disease Neuroimaging Initiative(ADNI),which contains 200 subjects of each group(AD,MCI and health control(HC)).In our work,the training set was divided into three categories,which are AD vs.MCI,AD vs.HC and MCI vs.HC.Then we propose to learn an end-to-end deep neural network classifier for MRI image recognition.This model was called MCINet.Lastly the lack number of training data yields over-fitting phenomena.We propose a method of a cross-modal transfer learning: from large-scale public data sets to MRI modality.Models pre-trained on a large-scale public dataset are used as initialization of network parameters and the MRI data set are used to adjust the parameters of some convolutional layers.From the experimental results,it can be seen that the method of this thesis is well solved for the case where the data set size is relatively small.We show that the proposed method achieves higher accuracy in the case of fewer labeled training samples.Finally,the application of different visualization methods is critical to understand CNN decisions,increase clinical impact,and trust in computer-based decision support systems.Although CNN provides good classification results,they are difficult to visualize and explain.However,in medical decision-making,it is important to explain the behavior of the machine learning model and to have medical experts verify the diagnosis.Therefore,this thesis detects Alzheimer’s disease 3D convolutional neural network,and then uses two different visualization methods for analysis.Interpret the classification decision of the network by highlighting the relevant areas in the input image.Based on the results of these visualizations to help clinicians make accurate diagnoses.