The application of deep learning to early detection and automatedclassification of Alzheimer's disease (AD) has recently gained considerableattention as rapid progress in neuroimaging techniques has generatedlarge-scale multimodal neuroimaging data. Here we systematically reviewedpublications, where deep learning approaches and neuroimaging data were usedfor diagnostic classification of AD. A PubMed and google scholar search wasperformed to find deep learning papers for AD published between January 2013and July 2018, which were reviewed, evaluated, and classified by algorithms andneuroimaging types, and findings were summarized. The diagnostic classificationof AD using deep learning approaches and neuroimaging data was examined in 16studies. The approach to combine traditional machine learning forclassification and stacked auto-encoder (SAE) for feature selection hasproduced accuracies of up to 98.8% for AD classification and 83.7% forprediction of conversion from mild cognitive impairment (MCI), a prodromalstage of AD, to AD. Deep learning approaches such as convolutional neuralnetwork (CNN) or recurrent neural network (RNN) using neuroimaging data withoutpreprocessing for feature selection have yielded accuracies of up to 96.0% forAD classification and 84.2% for MCI conversion prediction. Furthermore, thebest classification performance was obtained when multimodal neuroimaging dataas well as fluid biomarkers were integrated. Deep learning approaches withoutpreprocessing neuroimaging data for feature selection, a major bottleneck oftraditional machining learning in high-dimensional data, continue to improvetheir performance and to show great promise in the diagnostic classification ofAD using multimodal neuroimaging data.
translated by 谷歌翻译