Abstract
OBJECTIVES: Deep medullary veins support the venous drainage of the brain and may display abnormalities in the context of different cerebrovascular diseases. We present and evaluate a method to automatically detect and quantify deep medullary veins at 7 T. METHODS: Five participants were scanned twice, to assess the robustness and reproducibility of manual and automated vein detection. Additionally, the method was evaluated on 24 participants to demonstrate its application. Deep medullary veins were assessed within an automatically created region-of-interest around the lateral ventricles, defined such that all veins must intersect it. A combination of vesselness, tubular tracking, and hysteresis thresholding located individual veins, which were quantified by counting and computing (3-D) density maps. RESULTS: Visual assessment was time-consuming (2 h/scan), with an intra-/inter-observer agreement on absolute vein count of ICC = 0.76 and 0.60, respectively. The automated vein detection showed excellent inter-scan reproducibility before (ICC = 0.79) and after (ICC = 0.88) visually censoring false positives. It had a positive predictive value of 71.6 %. CONCLUSION: Imaging at 7 T allows visualization and quantification of deep medullary veins. The presented method offers fast and reliable automated assessment of deep medullary veins. KEY POINTS: • Deep medullary veins support the venous drainage of the brain • Abnormalities of these veins may indicate cerebrovascular disease and quantification is needed • Automated methods can achieve this and support human observers • The presented method provides robust and reproducible detection of veins • Intuitive quantification is provided via count and venous density maps.