OBJECTIVE
To test the hypothesis that multi-shell diffusion models improve the characterization of microstructural alterations in cerebral small vessel disease (SVD), we assessed associations with processing speed performance, longitudinal change and reproducibility of diffusion metrics.
METHODS
We included 50 sporadic and 59 genetically defined SVD patients (CADASIL) with cognitive testing and standardized 3T MRI, including multi-shell diffusion imaging. We applied the simple diffusion tensor imaging (DTI) model and 2 advanced models: diffusion kurtosis imaging (DKI) and neurite orientation dispersion and density imaging (NODDI). Linear regression and multivariable random forest regression (including conventional SVD markers) were used to determine associations between diffusion metrics and processing speed performance. The detection of short-term disease progression was assessed by linear mixed models in 49 sporadic SVD patients with longitudinal high-frequency imaging (in total 459 MRIs). Inter-site reproducibility was determined in 10 CADASIL patients scanned back-to-back on 2 different 3T MRI scanners.
RESULTS
Metrics from DKI showed the strongest associations with processing speed performance (R 2 up to 21%) and the largest added benefit on top of conventional SVD imaging markers in sporadic SVD and CADASIL patients with lower SVD burden. Several metrics from DTI and DKI performed similarly in detecting disease progression. Reproducibility was excellent (intraclass correlation coefficient >0.93) for DTI and DKI metrics. NODDI metrics were less reproducible.
CONCLUSION
Multi-shell diffusion imaging and DKI improve the detection and characterization of cognitively relevant microstructural white matter alterations in SVD. Excellent reproducibility of diffusion metrics endorses their use as SVD markers in research and clinical care. Our publicly available inter-site dataset facilitates future studies.
CLASSIFICATION OF EVIDENCE
This study provides Class I evidence that in patients with SVD, diffusion MRI metrics are associated with processing speed performance.