Abstract
Structural and metabolic connectivity are advanced features that facilitate the diagnosis of patients with Alzheimer’s disease (AD) and mild cognitive impairment (MCI). Connectivity from a single imaging modality, however, did not show evident discriminative value in predicting MCI-to-AD conversion, possibly because the inter-modal information was not considered when quantifying the relationship between brain regions. Here we introduce a novel approach that extracts connectivity based on both structural and metabolic information to improve AD early diagnosis. Principal component analysis was performed on each imaging modality to extract the key discriminative patterns of each brain region in an independent auxiliary domain composed of AD and normal control (NC) subjects, which were then used to project the two subtypes of MCI to the low-dimensional space. The connectivity between each target brain region and all other regions was quantified via a multi-task regression model using the projected data. The prediction performance was evaluated in 75 stable MCI (sMCI) patients and 51 progressive MCI (pMCI) patients who converted to AD within 3 years. We achieved 79.37% accuracy, with 74.51% sensitivity and 82.67% specificity, in predicting MCI-to-AD progression, superior to other existing algorithms using either structural and metabolic connectivities alone or a combination thereof. Our results demonstrate the effectiveness of multi-modal connectivity, serving as robust biomarker for early AD diagnosis.
Highlights
Alzheimer’s disease (AD) is the most common neurodegenerative disease characterized by shortterm memory loss and a decline of cognitive functions, including executive, visuospatial abilities, and language (Braa and Braak, 1991)
Brain network analysis is an efficient tool in characterizing topological organization of the brain, which has been widely used in investigating cerebral abnormalities caused by mental disorders, such as AD and Mild cognitive impairment (MCI) (Tijms et al, 2013; Kong et al, 2015; Kim et al, 2016; Wang et al, 2016)
We examined the diagnostic performance of the multi-modal connectivity (MMC) by cross-validating the results with a support vector machine (SVM) (Vapnik, 2000)
Summary
Alzheimer’s disease (AD) is the most common neurodegenerative disease characterized by shortterm memory loss and a decline of cognitive functions, including executive, visuospatial abilities, and language (Braa and Braak, 1991). At the brain parcel level, Pearson correlation, Euclidean distance, mutual information, and the Kullback–Leibler divergence of probability density function were, respectively, used to measure the relationship between the properties (e.g., cortical thickness, volume and metabolism) of distinct brain regions (Wee et al, 2013; Kong et al, 2014; Raamana et al, 2015; Zheng et al, 2015; Jiang et al, 2017; Li et al, 2017; Liu et al, 2017); at the voxel-level, Pearson correlations between gray matter density of small patches consisting of a serial of voxels (e.g., 3 × 3 × 3 voxels in each patch) were used to construct the covariance matrix (Tijms et al, 2012) These networks were reported with “small-world” organization and altered network matrices with the progression of AD (Tijms et al, 2013; Kong et al, 2015; Kim et al, 2016; Wang et al, 2016); and achieved an evident performance superior to original anatomical and metabolic features in classifying AD and MCI cohorts from the NCs (Wee et al, 2013; Liu M. et al, 2014; Zheng et al, 2015; Yao et al, 2016; Zhao et al, 2017)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
