Brain Artery Segmentation for Structural MRI

Bertram Sabrowsky-Hirsch, Ahmed AlShenoudy, Stefan Thumfart, Michael Giretzlehner, Josef Scharinger

Keywords: artificial intelligence, segmentation, brain arteries, sMRI

Abstract: The visualization of brain arteries in neuroimaging scans is essential for evaluating neurological disorders effectively. In this paper, we propose a deep learning-based method for the segmentation of brain arteries in structural MR images (sMRI), where their delineation poses a significant challenge due to the lack of contrast. Our fully automated strategy leverages two modules: one for generating pseudo labels from angiographic MR images and another for pairing these labels with five distinct sMRI sequences. The process enables us to construct a large dataset of 2626 labelled images for 669 patients used to train our segmentation model. In our experiments, our model achieved an average Dice Similarity Coefficient (DSC) of 0.66 across all sMRI around the central Circle of Willis structure in a 5-fold cross validation. We outline our results for each evaluated sMRI sequence, out of which we identify PD with a DSC of 0.7 as the best alternative to angiographic images.

Submission Number: 41