Hybrid multivariate pattern analysis combined with extreme learning machine for Alzheimer's dementia diagnosis using multi-measure rs-fMRI spatial patterns

Abstract

Background: Early diagnosis of Alzheimer's disease (AD) and Mild Cognitive Impairment (MCI) is essential for timely treatment. Machine learning and multivariate pattern analysis (MVPA) for the diagnosis of brain disorders are explicitly attracting attention in the neuroimaging community. In this paper, we propose a voxel-wise discriminative framework applied to multi-measure resting-state fMRI (rs-fMRI) that integrates hybrid MVPA and extreme learning machine (ELM) for the automated discrimination of AD and MCI from the cognitive normal (CN) state.

Materials and methods: We used two rs-fMRI cohorts: the public Alzheimer's disease Neuroimaging Initiative database (ADNI2) and an in-house Alzheimer's disease cohort from South Korea, both including individuals with AD, MCI, and normal controls. After extracting three-dimensional (3-D) patterns measuring regional coherence and functional connectivity during the resting state, we performed univariate statistical t-tests to generate a 3-D mask that retained only voxels showing significant changes. Given the initial univariate features, to enhance discriminative patterns, we implemented MVPA feature reduction using support vector machine-recursive feature elimination (SVM-RFE), and least absolute shrinkage and selection operator (LASSO), in combination with the univariate t-test. Classifications were performed by an ELM, and its efficiency was compared to linear and nonlinear (radial basis function) SVMs.

Results: The maximal accuracies achieved by the method in the ADNI2 cohort were 98.86% (p<0.001) and 98.57% (p<0.001) for AD and MCI vs. CN, respectively. In the in-house cohort, the same accuracies were 98.70% (p<0.001) and 94.16% (p<0.001).

Conclusion: From a clinical perspective, combining extreme learning machine and hybrid MVPA applied on concatenations of multiple rs-fMRI biomarkers can potentially assist the clinicians in AD and MCI diagnosis.

Fig 1. Descriptions of the proposed framework in this study.

Block (a) presents the 3-D feature measure extractions from preprocessed fMRI scans. Block (b) describes the LOO-CV and 10-fold-CV cross validation for ADNI2 and in-house cohorts, respectively. Block (c) presents the multivariate feature reduction techniques using LASSO and SVM-RFE. The combined univariate t-test and multivariate LASSO as well as SVM-RFE informative features are trained by ELM and SVM classifiers as illustrated in block (d). Finally, the trained classifiers and testing features are used to evaluate the performance as in block (e).

Fig 2

An example of one-fold univariate statistical two-sample t-test on ReHo maps between two training analytical groups, i.e., AD against CN (left subfigure) and MCI against CN (right subfigure). The threshold was set to p-value<0.05 with cluster size of 85 voxels (2295 mm³), which corresponded to a corrected p-value<0.05. The t-test maps are overlaid on the anatomical image. The hot and cold colours represent positive and negative changes.

Fig 3. Illustration of the hybrid combination of univariate t-test and MVPA feature reduction techniques (SVM-RFE and LASSO) on the 3-D cross-validated fMRI measures.

Fig 4. An example of cross-validated MSE of LASSO fit with a parameter lambda (λ).

Fig 5. Univariate t-statistical difference maps between AD and CN groups of ten measures extracted from in-house cohort.

Voxels with p-value<0.05 and cluster size of 85 voxels (2295 mm³) corresponding to a corrected p-value<0.05 were used to identify the significant clusters. Hot and cold colours indicate AD-related measures increases and decreases, respectively.

Fig 6. Univariate t-statistical difference maps between MCI and CN groups of ten measures extracted from in-house cohort.

Voxels with p-value<0.05 and cluster size of 85 voxels (2295 mm³) corresponding to a corrected p-value<0.05 were used to identify the significant clusters. Hot and cold colours indicate MCI-related measures increases and decreases, respectively.