Author: Tina Ehtiati, Grace Jianan Gang, Limei Ma, Oleg Shekhtman, Visish M. Srinivasan 👨🔬
Affiliation: Siemens Medical Solutions USA, Inc., University of Pennsylvania 🌍
Purpose: Saccular aneurysms are the most common type of intracranial aneurysm and are typically treated by endovascular embolization. The procedure requires approximately orthogonal fluoroscopy images at two appropriate working angles from biplane C-arms to guide device placement and ensure parent vessel patency. Currently, the angles are manually selected based on an intra-operative 3D-DSA, which can be time-consuming due to anatomical complexities and mechanical constraints of the system. We aim to develop an automatic angle prediction model to improve the efficiency of the procedure.
Methods: We first segmented the aneurysm from its parent vessel using a deep learning model. Selection criteria for working angles adopted by neurosurgeons were then abstracted into mathematical rules for prediction. For the neck view (lateral view of the aneurysm), the angle was selected using a weighted voting method based on three criteria: maximizing neck diameter, maximizing projection area of parent vessels, and minimizing the overlap between aneurysm and parent vessels. For the barrel view (along the longitudinal axis of parent vessels), we developed a method to automatically compute such direction from centerline of the parent/branch vessels. The method was evaluated on 85 aneurysm models from a public database and compared with those manually annotated by a panel of experienced neurosurgeons.
Results: For neck view, the absolute angle difference between prediction and ground truth is within 10o for 41%, 20o for 75%, and 30o for 89% of cases; for barrel view, that are 52%, 80%, and 95%. The complexity of vessel structures is highly correlated with prediction performance. All aneurysms cases with simple, non-branching parent vessels have angle differences <33o; 35% of aneurysm cases with branching vessels exceeds 20o.
Conclusion: This work presents an automatic approach for determining working angles during aneurysm treatment. The method holds significant potential for integration into clinical workflows to enhance procedural efficiency.