Early Diagnosis Of Alzheimer's Disease Through Transfer Learning

Alzheimer's disease(AD)is a progressive degenerative disease that directly affects mental functions and disturbs daily routine life.AD is the most dominant kind of dementia.The detection of AD at an early stage is a challenging task for practitioners or researchers.Machine and deep learning-based approaches are readily available to solve various problems related to AD classification.For accurate classification of dementia stages,we need highly discriminative features for evaluation.Mild cognitive impairment(MCI)detection using magnetic resonance image(MRI),plays a crucial role in treating dementia disease at an early stage.Algorithms require a large number of annotated datasets for training the classification model.However,due to fewer annotated datasets,over-fitting problems hinder the performance of different approaches.There is no proper cure for AD,but correct classification findings are beneficial for certain therapies that slow down the disease rate.In this study,we overcame these issues by using layer-wise transfer learning and tissue segmentation of brain images to diagnose the early stage of AD.we used the VGG architecture family with pre-trained weights that increased the classification performance.We proposed the model segregates between normal control(NC),the early mild cognitive impairment(EMCI),the late mild cognitive impairment(LMCI),and the AD.The samples attained from the Alzheimer's Disease Neuroimaging Initiative(ADNI)database.Tissue segmentation was applied to each subject to extract the gray matter(GM)tissue.The proposed method applied on the preprocessing data has achieved the highest performance.The classification accuracy rates obtained on AD vs NC is 98.73%.It is used to distinguish between EMCI vs LMCI patients' with a testing accuracy of 83.72%.In contrast,the remaining classes accuracy is more than 80%.Furthermore,we proposed a Siamese convolutional neural network(SCNN)approach inspired by VGG16(also called Oxford Net)to classify dementia stages.Through this approach,we attained an excellent test accuracy of 99.05%.Furthermore,we proposed another two transfer learning strategies with augmentation for normal control(NC)classification,mild cognitive impairment(MCI),and AD.The proposed models obtained the state-of-the-art performance in terms of accuracy for three binary classification problems such as NC vs.AD,NC vs.MCI,and MCI vs.AD.We achieved 97.68%,94.25%,and 92.18% respectively.Finally,we provide a comparative analysis with other studies which shows that the proposed model outperformed the state-of-the-art models in terms of classification accuracy.