Abstract
This paper presents a semi-automated software suite designed for the quantitative analysis of vascular structures during medical interventions. By integrating efficient segmentation, skeletonization-based navigation, and pattern-matching algorithms such as Dynamic Time Warping (DTW), the system facilitates the rapid identification and measurement of stenoses and aneurysms. The workflow is optimized for real-time clinical use, significantly reducing manual effort while maintaining high accuracy.
Introduction
During interventional procedures, such as stent placement or aneurysm coiling, surgeons require precise 3D maps of the vascular system. Traditional methods for identifying pathological segments and calculating dimensions (length, diameter, volume) are often time-consuming and labor-intensive, creating bottlenecks in high-pressure clinical environments. This work addresses the need for a highly automated "one-click" workflow capable of processing complex Digital Subtraction Angiography (DSA) images fast enough for intraoperative use.
Methods
The proposed system employs a multi-stage workflow to streamline vascular analysis:
Segmentation and Skeletonization
Vessels are isolated using a region-growing algorithm with thresholding. A distance transformation is then applied to extract a "skeleton"—a set of centerlines representing the vascular tree.
Intelligent Navigation and Pruning
To simplify navigation through complex structures, the system allows users to adjust the level of detail ($L$). Short, insignificant branches are pruned based on this threshold, leaving a clean path for navigation via simple controls.
Quantification and Detection
When a user selects a vessel, a 3D pointer snaps to the skeleton path. The system automatically generates 2D cross-sections and uses iterative line-fitting to determine minimum diameter and narrowing length. For automatic detection, Dynamic Time Warping (DTW) compares vessel diameter profiles against reference patterns to flag potential stenoses. For aneurysms, a Volume of Interest (VOI) approach is used, filling "intensity holes" caused by turbulent flow to ensure accurate volume calculation for embolization planning.
Results and Evaluation
The system demonstrates high performance, processing a $256 \times 256 \times 300$ volume in under 35 seconds on standard hardware. Quantitative evaluation using a physical phantom model showed a volume measurement error of approximately 10%, which is clinically acceptable for treatment planning.
The software was successfully integrated into the Siemens "Syngo" platform and achieved FDA compliance. While DTW-based stenosis detection is effective in larger vessels, it currently faces challenges with over-detection in thin, noisy peripheral vessels.
Conclusion
The presented software suite offers a robust solution for intraoperative vascular analysis, transforming skeletonization from a mathematical abstraction into an intuitive navigation tool. Future work will focus on refining diameter calculations and reducing false positives in automatic detection to further minimize the need for manual confirmation.
Summary updated