Deep Learning in Alzheimer's Disease: Diagnostic Classification and Prognostic Prediction Using Neuroimaging Data

Abstract

Deep learning, a state-of-the-art machine learning approach, has shown outstanding performance over traditional machine learning in identifying intricate structures in complex high-dimensional data, especially in the domain of computer vision. The application of deep learning to early detection and automated classification of Alzheimer's disease (AD) has recently gained considerable attention, as rapid progress in neuroimaging techniques has generated large-scale multimodal neuroimaging data. A systematic review of publications using deep learning approaches and neuroimaging data for diagnostic classification of AD was performed. A PubMed and Google Scholar search was used to identify deep learning papers on AD published between January 2013 and July 2018. These papers were reviewed, evaluated, and classified by algorithm and neuroimaging type, and the findings were summarized. Of 16 studies meeting full inclusion criteria, 4 used a combination of deep learning and traditional machine learning approaches, and 12 used only deep learning approaches. The combination of traditional machine learning for classification and stacked auto-encoder (SAE) for feature selection produced accuracies of up to 98.8% for AD classification and 83.7% for prediction of conversion from mild cognitive impairment (MCI), a prodromal stage of AD, to AD. Deep learning approaches, such as convolutional neural network (CNN) or recurrent neural network (RNN), that use neuroimaging data without pre-processing for feature selection have yielded accuracies of up to 96.0% for AD classification and 84.2% for MCI conversion prediction. The best classification performance was obtained when multimodal neuroimaging and fluid biomarkers were combined. Deep learning approaches continue to improve in performance and appear to hold promise for diagnostic classification of AD using multimodal neuroimaging data. AD research that uses deep learning is still evolving, improving performance by incorporating additional hybrid data types, such as-omics data, increasing transparency with explainable approaches that add knowledge of specific disease-related features and mechanisms.

Keywords: Alzheimer's disease; artificial intelligence; classification; deep learning; machine learning; magnetic resonance imaging; neuroimaging; positron emission tomography.

Figure 1

The multilayer perceptron procedure. After the initial error value is calculated from the given random weight by the least squares method, the weights are updated by a back-propagation algorithm until the differential value becomes 0.

Figure 2

Common activation functions used in deep learning (red) and their derivatives (blue). When the sigmoid is differentiated, the maximum value is 0.25, which becomes closer to 0 when it continues to multiply.

Figure 3

Architectural structures in deep learning: (A) RBM (Hinton and Salakhutdinov, 2006) (B) DBM (Salakhutdinov and Larochelle, 2010) (C) DBN (Bengio, 2009) (D) CNN (Krizhevsky et al., 2012) (E) AE (Fukushima, ; Krizhevsky and Hinton, 2011) (F) Sparse AE (Vincent et al., 2008, 2010) (G) Stacked AE (Larochelle et al., ; Makhzani and Frey, 2015). RBM, Restricted Boltzmann Machine; DBM, Deep Boltzmann Machine; DBN, Deep Belief Network; CNN, Convolutional Neural Network; AE, Auto-Encoders.

Figure 4

PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) Flow Chart. From a total of 389 hits on Google scholar and PubMed search, 16 articles were included in the systematic review.

Figure 5

Comparison of diagnostic classification accuracy of pure deep learning and hybrid approach. Four studies (gray) have used hybrid methods that combine deep learning for feature selection from neuroimaging data and traditional machine learning, such as the SVM as a classifier. Twelve studies (blue) have used deep learning method with softmax classifier for diagnostic classification and/or prediction of MCI to AD conversion. (A) Accuracy comparison between articles. (B) Number of studies published per year. (C) Average classification accuracy of each methods.

Figure 6

Changes in accuracy by types of image resource. MRI scans were used in 13 studies, FDG-PET scans in 10, both MRI and FDG-PET scans in 12, and both amyloid PET and FDG-PET scans in 1. The performance in AD/CN classification yielded better results in PET data compared to MRI. Two or more multimodal neuroimaging data types produced higher accuracies than a single neuroimaging technique.