Use of a Sparse-Response Deep Belief Network and Extreme Learning Machine to Discriminate Alzheimer's Disease, Mild Cognitive Impairment, and Normal Controls Based on Amyloid PET/MRI Images

Abstract

In recent years, interest has grown in using computer-aided diagnosis (CAD) for Alzheimer's disease (AD) and its prodromal stage, mild cognitive impairment (MCI). However, existing CAD technologies often overfit data and have poor generalizability. In this study, we proposed a sparse-response deep belief network (SR-DBN) model based on rate distortion (RD) theory and an extreme learning machine (ELM) model to distinguish AD, MCI, and normal controls (NC). We used [18F]-AV45 positron emission computed tomography (PET) and magnetic resonance imaging (MRI) images from 340 subjects enrolled in the ADNI database, including 116 AD, 82 MCI, and 142 NC subjects. The model was evaluated using five-fold cross-validation. In the whole model, fast principal component analysis (PCA) served as a dimension reduction algorithm. An SR-DBN extracted features from the images, and an ELM obtained the classification. Furthermore, to evaluate the effectiveness of our method, we performed comparative trials. In contrast experiment 1, the ELM was replaced by a support vector machine (SVM). Contrast experiment 2 adopted DBN without sparsity. Contrast experiment 3 consisted of fast PCA and an ELM. Contrast experiment 4 used a classic convolutional neural network (CNN) to classify AD. Accuracy, sensitivity, specificity, and area under the curve (AUC) were examined to validate the results. Our model achieved 91.68% accuracy, 95.47% sensitivity, 86.68% specificity, and an AUC of 0.87 separating between AD and NC groups; 87.25% accuracy, 79.74% sensitivity, 91.58% specificity, and an AUC of 0.79 separating MCI and NC groups; and 80.35% accuracy, 85.65% sensitivity, 72.98% specificity, and an AUC of 0.71 separating AD and MCI groups, which gave better classification than other models assessed.

Deep Learning with Neuroimaging and Genomics in Alzheimer’s Disease

“…As described in the following subsections, there are some studies for AD research with genomics data using various deep learning models (Table 1), including the prediction of AD risk, the prediction of AD-specific nucleotide alteration sites (i.e., splicing sites), and the prediction of the virtual disease/molecular progress of AD. [41] FNNs Gene expression Predict AD risk Park, J. et al [42] GANs Gene expression Predict the virtual disease/molecular progress of AD Kim et al [43] Residual CNNs Gene expression Predict AD-specific nucleotide alteration sites (i.e., splicing sites) Park, C. et al [44] FNNs Gene expression, DNA methylation Predict AD risk Ju et al [45] Autoencoders MRI Predict early diagnosis of AD Shen et al [46] DBNs PET Distinguish AD from MCI Zhou, P. et al [47] Sparse-response DBNs PET, MRI Predict AD risk Ning et al [48] FNNs SNPs, MRI (brain measures) Predict AD risk Zhou, T. et al [49] Three-stage FNNs SNPs, ROIs in PET, ROIs in MRI Predict AD risk Zhou, J. et al [50] CNNs SNPs, ROIs in MRI Predict AD risk…”

“…Zhou, P. et al [ 47 ] utilized a deep learning-based approach to predict AD using PET and MRI images. Their deep learning approach was characterized by sparse-response DBNs [ 39 ] (see Section 3.2.5 ), which was used for extracting features from the images.…”

“…Zhou, P. et al [ 47 ] indicated that the proposed approach (AUC = 0.87) outperformed the benchmarking models, such as CNNs (AUC = 0.77) and DBNs (AUC = 0.83) for distinguishing AD and normal controls. In addition, the proposed approach (AUC = 0.79) exceeded CNNs (AUC = 0.60) and DBNs (AUC = 0.73) for distinguishing mild cognitive impairment and normal controls [ 47 ]. Moreover, the proposed approach (AUC = 0.71) surpassed CNNs (AUC = 0.60) and DBNs (AUC = 0.68) for distinguishing AD and mild cognitive impairment [ 47 ].…”

“…In addition, the proposed approach (AUC = 0.79) exceeded CNNs (AUC = 0.60) and DBNs (AUC = 0.73) for distinguishing mild cognitive impairment and normal controls [ 47 ]. Moreover, the proposed approach (AUC = 0.71) surpassed CNNs (AUC = 0.60) and DBNs (AUC = 0.68) for distinguishing AD and mild cognitive impairment [ 47 ].…”

“…The main weakness of the study by Zhou, P. et al [ 47 ] is that other deep learning algorithms such as GANs and other machine learning algorithms such as random forests were not utilized for comparison.…”