Deep Learning based Vessel Suppression on Contrast Enhanced Brain MRI Images

Motivation: Contrast-enhanced Gradient Echo (GRE) sequences offer good lesion visualization but enhancing blood vessels hinder delineation from actual pathology. Goal(s): Develop a retrospective post-processing algorithm to perform vessel suppression Approach: We synthesize a spin echo (SE) sequence from GRE sequences using a deep learning (DL) network and use the difference image to create a vessel map. Using the vessel map, the vessel signals are suppressed in the input GRE image. Results: Quantitative and qualitative results show that the vessel suppressed images have the same contrast enhancement inside lesions, and reduced contrast enhancement in the vessel signals when compared to non-vessel suppressed GRE images. Impact: The proposed method paves the way for a novel post-processing way of achieving vessel suppression without having to rescan with specialized imaging protocols. This algorithm will be impactful for cases with small lesions like metastases.