Diffusion MRI metrics of brain microstructure in Alzheimer’s disease: Boosting disease sensitivity with multi‐shell imaging and advanced pre‐processing

Abstract Background Diffusion MRI (dMRI) provides novel measures of white matter (WM) microstructure, offering biomarkers that provide insight into brain aging and Alzheimer’s disease. As advanced acquisitions are implemented (multi‐shell/multi‐band; MB), novel models are developed to more accurately represent tissue microstructure. In its third phase, the Alzheimer’s Disease Neuroimaging Initiative (ADNI‐3) updated their acquisition protocols to include MB dMRI. As acquisition protocols are being updated, it is important to understand which processing steps consistently improve dMRI model fit and derived measure sensitivity to biological effects. Here, we compared state‐of‐the‐art processing methods for ADNI‐3 MB scans to (1) establish an updated dMRI ADNI processing pipeline, and (2) fit more sophisticated tissue parameters sensitive to clinical measures of burden. Method Baseline MB ADNI‐3 scans (N=137; Table 1) were downloaded and processed with both the dMRI pipeline we established during ADNI‐2 (Pipeline‐A) and the one we have updated during ADNI‐3 (Pipeline‐B; Fig. 1). We obtained scalar diffusion tensor imaging (DTI) maps from the down‐sampled preprocessed dMRI, and higher model metrics using all diffusion volumes using NODDI. Covarying for age, sex, age*sex, and modeling acquisition site as a random effect, we tested associations between regions of interest (ROI; Table 2) diffusion scalar measures and a set of standardized clinical measures (Fig. 2). We modeled effects of age in controls, covarying for sex, age*sex and modeling site as a random effect. All results were corrected for multiple comparisons (FDR). Within WM, voxel‐wise DTI model error (RMSE) was compared across pipelines with paired t ‐tests in controls (Fig. 3). Result Fractional anisotropy measures were most affected by the different preprocessing methods, while diffusivity metrics were most strongly correlated with standard clinical measures of cognitive disease burden, such as CDR‐sob. Metrics from NODDI, such as ODI, behaved similarly to FA, showing greatest differences in effect sizes across pipelines. Conclusion We are currently applying Pipeline‐B to all available ADNI‐3 dMRI scans, to make the data available to the imaging community. As new preprocessing and denoising tools are available and acquisition protocols become more complex, our analyses indicate that preprocessing steps can sensitize diffusion metrics to differences in clinical disease burden.