Automatic Blood Vessel Segmentation using Magnitude Images from Magnetic Resonance Elastography

Quantifying hepatic stiffness through Magnetic Resonance Elastography (MRE) is essential for assessing liver health, yet it remains a complex task, due to the need for precise region of interest (ROI) placement. This process requires ROIs to be positioned in areas free from blood vessels as their presence can distort stiffness measurements. However, the poor quality of magnitude images often degraded by motion artefacts due to vibrations in the MRE process, makes manual placement challenging. To address this, our study aims to develop an automatic method for hepatic blood vessel detection using a Deep Learning (DL) model. We employed a 2D U-Net architecture for vessel segmentation, designed to differentiate blood vessels from liver tissue in MRE magnitude images affected by artefacts. The accuracy of the 2D U-Net segmentation was quantitatively evaluated using two primary metrics: the Dice Similarity Coefficient (DSC) and the Hausdorff Distance (HD). Our proposed algorithm achieved high performance, with a high DSC indicating reliable segmentation overlap and a low HD reflecting precise boundary delineation. This AI based method demonstrates a great potential for improving the accuracy of hepatic stiffness quantification, providing a consistent and efficient alternative to manual vessel exclusion. By ensuring precise blood vessel detection