Diffusion MRI indices of brain microstructure: Evaluating the Ball‐and‐Stick model in distinguishing Alzheimer’s disease

Abstract

AbstractBackgroundDiffusion MRI is a valuable tool to assess white matter (WM) microstructure biomarkers of brain aging and Alzheimer’s disease (AD). The most common model used to extract diffusion properties is the diffusion tensor model (DTI). However, DTI cannot resolve complex WM microstructure and suffers from partial volume effects. Further, the majority of multi‐site imaging clinical studies, such as ADNI, are limited to single‐shell acquisitions due to time constraints. Therefore, multi‐compartment models are not feasible to overcome these limitations except multi‐tensor models, such as Ball‐and‐Stick (B&S) under some constraints. B&S offers a novel approach to derive biomarkers for early detection of AD because it splits the diffusion‐weighted MR signal into several anisotropic components (Stick) and a single isotropic component (Ball). This model is implemented in Anima software. Here, we propose a classification approach using four DTI and two B&S indices quantified in various anatomical regions of interest (ROIs) to assess and compare their ability to distinguish AD.MethodCross‐sectional data were available for 413 ADNI‐3 participants: 68 AD (age: 76.81±8.3; 50% women) and 345 cognitively normal (CN; age:72.91±7.27; 62% women). We processed T1‐weighted and diffusion sequences following the pipeline in Fig.1. We obtained four DTI maps: fractional anisotropy (FA), axial (AxD), mean (MD), radial (RD) diffusivity; and two B&S maps: free water weight (FWW) and Stick‐axial diffusivity (ST). Diffusion maps were quantified within a set of anatomical ROIs. Then, we performed a high‐dimensional pattern classification approach. For each metric, we consider all brain ROIs jointly and identify a minimal set of regions whose values jointly maximally differentiate between AD and CN.ResultB&S model with regional FWW and ST measures shows the greatest discrimination between AD and CN (FWW ROCAUC = 0.942; ST ROCAUC = 0.938) compared to DTI (Table 1). Left PHG Parahippocampal Gyrus and Right Hippocampus regions have the highest contribution in the classification (Table 2, Fig.2).ConclusionOur results suggest B&S measures are more sensitive to discriminate AD from CN than DTI measures, widely used in clinical research. B&S has the advantage of identifying and separating crossing fibers, and therefore, has the potential to offer greater and novel insights into the underlying pathology of AD.