Evaluating NODDI‐based biomarkers of Alzheimer's disease: Neuroimaging / Optimal neuroimaging measures for early detection. (7th December 2020)

• Record Type: Journal Article
• Authors: Reina, Julio E Villalon; Nir, Talia M; Thomopoulos, Sophia I; Salminen, Lauren; Jahanshad, Neda; Thompson, Paul
• 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.042297
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: Multi‐compartment (MC) diffusion MRI (dMRI) models are increasingly used to evaluate brain microstructure beyond the classic diffusion tensor model (DTI). The ability to separate the neural tissue into distinct compartments, i.e., intracellular volume fraction (ICVF), extracellular volume fraction (ECVF) and free‐water (FW), offers novel biomarkers for early detection of cognitive decline and Alzheimer's disease. The most commonly used MC model used is Neurite Orientation Dispersion and Density Imaging (NODDI). Recent evidence shows that the voxel‐wise volume fractions compartment can be wrongly estimated if the same T2 relaxation time is assumed for the gray matter (GM), white matter (WM) and corticospinal fluid (CSF). To more accurately estimate the tissue compartments, here we use a novel method recently proposed by the DMIPY package (https://github.com/AthenaEPI/dmipy), that initially estimates a response function for each brain region (GM, WM and CSF) and uses these to better estimate the ICVF and FW compartments. Method: We analyzed ADNI‐3 multi‐shell diffusion MRI (dMRI) data to estimate NODDI measures and to assess their ability to predict mild cognitive impairment (MCI). For each measure, we used regularized logistic regression to find cohesive clusters of brain tissue that contribute to correct classification. Data included 34 participants with MCI (mean age: 75.6±7.8yrs, 22M/9F) and 74 cognitively normal individuals (CN; mean age= 74.1±7.1yrs; 25M/39F). Siemens Prisma 3T multi‐shell dMRI data included 13 b0 images, 6 b=500, 48 b=1000, and 60 b=2000 s/mm2 diffusion weighted 32 images (voxel size=2mm, TE=71ms, TR=3300ms, δ=13.6ms, Δ=35ms). We fit three types of NODDI: the classic NODDI model, the NODDI model with a GM tissue response function, and with a WM response function. Result: The NODDI model with the GM response function increased the accuracy of the logistic regression classification (see Table 1). The highest recall was achieved by the neurite dispersion index (ODI) in all three models. Conclusion: Adding specific response functions for the GM and WM in NODDI improved classification accuracies and recall of MCI subjects. The distribution of the classifying regions was different across the three types of NODDI (Figures 1, 2 and 3).