Extracting Circle of Willis Topology Using Deep Learning
Author Information
Author(s): Dieuwertje Alblas, Iris N. Vos, Micha M. Lipplaa, Christoph Brune, Irene C. van der Schaaf, Mireille R. E. Velthuis, Birgitta K. Velthuis, Hugo J. Kuijf, Ynte M. Ruigrok, Jelmer M. Wolterink
Primary Institution: University of Twente
Hypothesis
Can deep learning improve the automatic extraction of the Circle of Willis topology from 3D TOF-MRA images?
Conclusion
The proposed method successfully extracts personalized Circle of Willis topology with high accuracy, particularly for hypoplastic arteries.
Supporting Evidence
- The method achieved an average F1 score of 0.91 for classifying artery segments.
- Bifurcation points were detected with a median accuracy below 1.9 mm.
- The proposed framework can be customized for various downstream analysis approaches.
Takeaway
This study shows how computers can help doctors see the blood vessels in the brain better, especially when some are smaller than usual.
Methodology
The study used a deep learning model to extract the Circle of Willis topology from 3D TOF-MRA scans, employing a path optimization algorithm based on local artery orientations.
Potential Biases
Potential bias in the classification of artery segments due to inaccuracies in path extraction.
Limitations
The method may struggle with aplastic arteries and lacks validation for bifurcation points in the absence of arteries.
Participant Demographics
Healthy individuals from a research screening cohort.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website