Vessel segmentation from quantitative susceptibility maps for local oxygenation venography

Recent works have demonstrated that oxygenation in the brain can be measured via susceptibility shifts between vessels and neighboring regions in magnetic resonance images. To obtain a rich picture of local oxygenation, small venous vessels across the cortex need to be extracted. This work presents a novel vessel filter for the segmentation of vasculature in high resolution quantitative susceptibility mapping. The filter recursively estimates image ridges, resulting in highly specific responses. A global probability diffusion scheme further enhances the response along connected vessels while removing inconsistent ones. Vessel diameters are then estimated in order to refine the segmentation and model partial volume effects.