Diagnostic Classification and Prediction Using Neuroimaging Data Deep Learning in Neuro Degenerative Disease

Abstract

Deep learning has demonstrated excellent performance in finding complicated structures in the difficult high-dimensional data field, particularly in the field of computer vision, above classical machine learning. Recently, much consideration has been given to the application of extensive learning for early detection and automated categorization of Alzheimer disease (AD), as rapid advancement in neuroimaging technologies has created broad multimodal neuroimagery data. A thorough examination of articles was conducted using profound learning and neuroimaging data for the diagnostic classification of AD. A search for deep learning publications published between January 2013 to Juli 2018 in PubMed and Google Scholar was used to identify AD. These documents were analysed, appraised, categorised, and the conclusions were summarised by algorithm and neuroimaging type. Out of 16 research that met criteria of full inclusion, 4 employed a combination from profound learning and traditionals, and 12 used exclusively profound learning methods. The combination of classical machine classification learning and stacked autoencoder (SAE) for feature selection produated accuracy up to 98.8% for the AD classification and 83.7% for the prodromal stage of AD to AD conversion prediction. Deep learning techniques, such as the CNN or the Recent Neural Network (RNN), using neuroimaging data without pre-processing for feature selection have resulted in accurate AD classification of up to 96.0 percent and MCI conversion forecast of up to 84.2 percent. When multimodal neuroimaging and fluid biomarkers were coupled the best classification performance was achieved.