Diffusion structural network metrics as a predictor for Alzheimer's disease diagnosis in Down syndrome using support vector machine classification: Neuroimaging / Optimal neuroimaging measures for early detection. (7th December 2020)

• Record Type: Journal Article
• Authors: Brown, Stephanie SG; Mak, Elijah; Grigorova, Monika; Beresford‐Webb, Jessica; Walpert, Madeleine Jane; Jones, Elizabeth; Hong, Young T; Fryer, Tim D; Coles, Jonathan P; Aigbirhio, Franklin I; Handen, Benjamin L; Christian, Brad T; Klunk, William E; Menon, David K; Nestor, Peter J; Holland, Anthony J; Zaman, Shahid
• Is Part Of: Alzheimer's & dementia. Volume 16(2020)Supplement 5
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-12-07
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.040261
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: Alzheimer's Disease (AD) is known to pathologically develop in Downs Syndrome in almost one hundred percent of cases, although at present structural changes to the brain are minimally utilized for diagnoses of neurodegeneration due to limited understanding of which white matter alterations are directly relevant to clinical presentation. By novel application of machine learning to metrics of the structural connectome in Downs Syndrome, this project aims to identify specific white matter networks that significantly impact classification of cases into AD positive or negative. Method: A pilot sample of 21 healthy Down's syndrome participants and 4 Down's syndrome participants with a diagnostic consensus of either mild cognitive impairment (MCI) or dementia underwent diffusion‐weighted and T1‐weighted MRI scanning at 3T. These data form part of a consortium, which will be utilised to increase sample size for future iterations of this analysis. Whole brain structural connectomes were produced for each participant, utilizing fibre orientation distributions and spherical deconvolution informed filtering of the tractograms. For accurate localisation of nodes to brain regions, the T1‐weighted scan was pre‐processed, segmented, parcellated and co‐registered to diffusion space. Networks were created by examining connection density between T1‐delineated structural nodes. Connection density matrices were used to train and test a support vector machine (SVM) classifier. Results: Preliminary results show widespread reduction in network structural connectivity in the Down's syndrome participants diagnosed with dementia or mild cognitive impairment (Fig. 1). The parietal, occipital and to a lesser extent the temporal areas of the network exhibit the most marked decreases in connection density, and no edges of the network displayed increased connection density in dementia or MCI patients compared to healthy Downs syndrome. Multivariate analyses indicate that although SVM classification of the networks has a >80% accuracy, false‐negative classifications were common due to the low number of Down's syndrome dementia participants. Conclusions: This pilot application of SVM to diffusion‐derived brain networks in Down's syndrome demonstrates utility and feasibility as a diagnostic classifier. We expect that increased sample size should yield a more robust trained model.