Segmentation of Cerebral Arteries in 4D CT Perfusion: Effects of Temporal Data Augmentation and Skullstripping

Alena-Kathrin Golla, Lara-Jasmin Behrend, Andreas Ziebart, Anish Raj, Lothar R. Schad, Frank G. Zoellner, Sherif A. Mohamed

Keywords: Vessel Segmentation, Computed Tomography Perfusion, Data Augmentation

TL;DR: temporal data augmentation and skull stripping enhance PCT vessel segmentation

Abstract: Computed tomography perfusion (CTP) is used to quantitively evaluate the hemodynamics of the brain parenchyma in the assessment of cerebrovascular diseases like stroke or vasospams. We explore the segmentation of cerebral arteries from dynamic CTP on an initial data set of 20 patients. We use a 3D U-Net in combination with ratio-based sampling to extract the arterial vessel tree from 4D data. We investigate the benefit of temporal data augmentation and skull stripping. A shift of time frames by up to three time points in combination with skull stripping increased the DSC by 19.3% and decreased the ASSD by 38.2%.

Paper Type: both

Primary Subject Area: Segmentation

Secondary Subject Area: Application: Radiology

Paper Status: original work, not submitted yet

Source Code Url: This project is still in an early phase, we will consider code publication at a later point.

Data Set Url: clinical data not permitted for publication

Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.