Schauer J, Thamm F, Taubmann O, Maier A (2023)
Publication Type: Conference contribution, Original article
Publication year: 2023
Conference Proceedings Title: Bildverarbeitung für die Medizin 2023
DOI: 10.1007/978-3-658-41657-7_15
Non-contrast computed tomography (NCCT) is the primary first-line neuroimaging technique in the clinical workflow for patients with suspected ischemic stroke. We present a deep learning model to estimate the cerebral vessel tree from the NCCT instead of subsequently performed contrast-enhanced imaging techniques, e.g. computed tomography angiography (CTA). We employ a volumetric sliding window approach and feed the patches to a 3D U-Net. This U-Net has two outputs, a probability map that indicates vessel presence and a prediction of the corresponding CTA patch. The CTA regression target is used in addition to the supervised segmentation to optimize the 3D U-Net in a GAN-like manner in order to generate more realistic estimations for the vessel tree. Comparing our proposed model with the current state of the art for this task, a 2D U-Net operating on axial NCCT slices, we were able to slightly increase quantitative overlap metrics as well as achieve notably improved qualitative results w.r.t. spatial continuity of the segmented vessel tree.
APA:
Schauer, J., Thamm, F., Taubmann, O., & Maier, A. (2023). Cerebral Vessel Tree Estimation from Non-contrast CT using Deep Learning Methods. In Bildverarbeitung für die Medizin 2023.
MLA:
Schauer, Jonas, et al. "Cerebral Vessel Tree Estimation from Non-contrast CT using Deep Learning Methods." Proceedings of the Bildverarbeitung für die Medizin 2023 2023.
BibTeX: Download