Multiclass Diagnostic Framework for Cognitive problems using brain MRI: Alzheimer’s Disease perspective

Abstract

Alzheimer’s disease (AD), a long-lasting or progressive neurodegenerative brain disorder that deteriorates cognitive skills mostly in aged people and it is also sixth prominent cause of death worldwide. In clinical AD diagnosis, generally brain Magnetic Resonance Image (MRI) have acquired great significance due to their capability of retrieving significant features. The artificial intelligence based diagnostic algorithms such as deep learning techniques provide most promising results in diagnosis. The research article proposed a robust multi-class transfer learning framework for AD diagnosis with brain MRI data set acquired from both Open Access Series of Imaging Studies (OASIS) and Alzheimer’s Disease Neuroimaging Initiative (ADNI). A deep learning model Deep Convolutional Neural Network (DCNN) is constructed rigorously trained and validated for reference to prove its capability in grouping the highly correlated features from MRI slices. To obtain better performance, a Visual Geometry Group (VGG16) architecture is retrained with fourteen layers and tuned systematically to classify the subjects among the four categories viz. AD, low and stable (ls-MCI), progressive MCI (p-MCI) and Cognitive Normal (CN). This network achieved 99.17% average accuracy that prove the eminence of deep learning in AD diagnosis especially in MCI subject classification