Evaluating Segmentation Techniques for Circle of Willis in 4D Flow MRI: A Comparative Study
Jiaxin Zhang, Anouk S. Verschuur, Eric Mathew Schrauben, Mark Bakker, Pim van Ooij, Kyung Min Nam, Irene C. van der Schaaf, Chantal M. W. Tax · Proceedings on CD-ROM - International Society for Magnetic Resonance in Medicine. Scientific Meeting and Exhibition/Proceedings of the International Society for Magnetic Resonance in Medicine, Scientific Meeting and Exhibition · 2025
Motivation: 7T 4D flow magnetic resonance imaging (MRI) has enhanced the visualization and quantification of flow within cerebral arteries. However, differences in vessel segmentation can cause variability in flow quantification. Goal(s): This study aims to develop and validate automatic artery segmentation for 4D flow MRI, especially focusing on Circle of Willis (CoW). Approach: We compared the segmentation performance and flow measurements of two deep learning (DL) models (3D U-Net and nnUNet) a thresholding algorithm from QVT software with those of manual segmentation. Results: nnUNet demonstrates the best segmentation performance in extracting small vessels and QVT resulted in the highest flow estimates. Impact: Accurate automatic intracranial vessel segmentation methods decrease the need for manual intervention and facilitate the measurement of flow in smaller arteries.