Related Experiment Video
Updated: Apr 23, 2026
Novel and Innovative Hybrid Technique for Type A Aortic Dissection
Published on: March 28, 2025
Development and Validation of a Deep Learning-Based Segmentation Method for Fenestration Marker and Graft Body
Polycronis P Akouris1, Sangwook Kim2,3,4, Arshia P Javidan5,6
1Temerty Faculty of Medicine, University of Toronto, Toronto, ON, Canada.
Objective:
To develop and validate a deep learning-based segmentation method for accurate identification of fenestration markers and graft body contours in intraoperative fluoroscopic images during fenestrated endovascular aortic repair (FEVAR).
Methods:
A total of 500 intraoperative fluoroscopic images from 31 FEVAR procedures were manually annotated for fenestration markers and graft body contours. Using nnU-Net, a self-configuring deep learning framework, we developed SIGMA (Segmentation of fluoroscopic Images for Graft and Marker Analysis) to perform automatic segmentation. The data set was split into training (295 images, 18 procedures) and testing (205 images, 13 procedures) sets using a patient-wise separation. Model performance was evaluated using marker localization performance (MLP)-including sensitivity, specificity, and Euclidean distance-for spatial accuracy of fenestration marker detection, and Dice Similarity Coefficient (DSC) for segmentation accuracy.
Results:
In the test data set, SIGMA achieved a median Euclidean distance of 0.16 mm for fenestration marker localization, with sensitivity and specificity of 91.0% and 90.2%, respectively. The mean DSC was 79.5% for fenestration markers and 96.6% for graft body contours. Overall, the model demonstrated robust performance across data sets for both localization and segmentation tasks.
Conclusion:
This study demonstrates the feasibility of deep learning-based segmentation for real-time intraoperative analysis during FEVAR. Accurate identification of fenestration markers and graft body contours provides a foundation for advanced tools that can calculate fenestration angles and graft orientation from 2D fluoroscopy, offering practical, cost-effective intraoperative guidance.Clinical ImpactThe SIGMA deep learning segmentation model enables accurate identification of fenestration markers and graft body contours in intraoperative fluoroscopic images during FEVAR. This technology lays the foundation for real-time calculation of graft orientation and fenestration angles. By leveraging standard 2D fluoroscopy, SIGMA supports the development of intelligent overlay tools that enhance intraoperative precision, reduce cognitive workload, and improve procedural efficiency. With further refinement and integration, these tools may standardize outcomes across operators and institutions, offering a practical, scalable solution for improving accuracy in complex aortic interventions.
