Related Experiment Video
Updated: Aug 9, 2026
14:58
Comprehensive Endovascular and Open Surgical Management of Cerebral Arteriovenous Malformations
Published on: October 20, 2017
A generative adversarial framework for optimal view prediction in aneurysm embolization
Annekoos Schaap1, Irene C van der Schaaf2, Matthias Bechstein3
1Department of Electrical Engineering, Eindhoven University of Technology, Groene Loper 5, 5612AZ, Eindhoven, The Netherlands. a.schaap@tue.nl.
Summary
This study shows that generative adversarial networks can predict optimal C-arm views for endovascular coiling, improving safety and efficiency in aneurysm treatment. This AI approach reduces variability and radiation exposure for better patient outcomes.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Neurosurgery
Background:
- Optimal C-arm working views are crucial for safe and efficient endovascular coiling of intracranial aneurysms.
- Current manual C-arm angulation leads to operator variability and increased radiation exposure.
Purpose of the Study:
- To investigate the feasibility of a generative adversarial framework for predicting C-arm viewing directions from 3D rotational angiography (3DRA) data.
- To automate the selection of optimal C-arm working views, enhancing procedural safety and efficiency.
Main Methods:
- Developed a generator-discriminator framework predicting 3D view vectors from segmented vascular and aneurysm data.
- Tested three projection strategies (ray casting, DRR, MIP-OR) with CNN and U-Net generator architectures.
- Assessed performance using absolute dot product error and expert qualitative evaluation.
Main Results:
- The CNN generator with MIP-OR projections achieved the lowest mean absolute dot product error.
- Expert assessments indicated improved alignment with clinically acceptable working views.
- Combined geometric and expert-based evaluation is necessary for clinical usability.
Conclusions:
- Adversarial learning with anatomically informed projections shows promise for automated C-arm view prediction in neuroendovascular procedures.
- Future research will focus on larger datasets and procedural ground-truth views for enhanced reliability and generalizability.