A novel robust fully-automated approach to segment cerebral perivascular spaces on clinical T1-weighted images

Motivation: Quantitative measurements of cerebral perivascular spaces (PVS) are important for the assessment of small vessel disease and glymphatic system. Current methods for PVS lack inter-scanner reproducibility and generally require some degree of manual intervention. Goal(s): To present a novel, fully-automated algorithm which segments PVS on clinical T1-weighted images with excellent inter-scanner and test-retest reproducibility. Approach: The Frangi filter is employed to generate "vesselness" maps. Then, PVS are segmented using a threshold identified automatically based on the distribution of the non-zero "vesselness" values. Results: Estimated PVS count and mean diameter showed excellent intraclass correlation coefficients for inter-scanner reproducibility, inter-field-strength reproducibility, and test-retest repeatability. Impact: The simplicity and efficiency of our algorithm in robustly and automatically assessing PVS on clinical MRI facilitate its easy implementation in hospitals, clinical trials, and even retrospectively for MRI data already acquired, enabling multi-center, reproducible investigations of human PVS.