Source-linked AI summary
VOIDD: automatic vessel of intervention dynamic detection in PCI procedures
Ketan Bacchuwar, Jean Cousty, Régis Vaillant, Laurent Najman
TL;DR
PCI workflow monitoring is limited by contrast-agent exposure and ionizing radiation during X-ray-guided procedures. VOIDD combines contrast-enhanced vessel references with guidewire-navigation images to detect the intervention vessel and navigation sequence. Across 2213 images from 4 patients, it achieved 88%–92% vessel-of-intervention accuracy and 98%–99% accuracy when identifying frames without a guidewire tip.
Problem
PCI relies on X-ray guidance and contrast injection, creating limited contrast tolerance and ionizing-radiation exposure while motivating more precise procedure monitoring.
Method
VOIDD combines guidewire-tip segmentation, vessel-graph matching, and track assignment between contrast-enhanced reference images and navigation fluoroscopy.
Results
88%–92% vessel-of-intervention accuracy was obtained over 1513 navigation images, while guidewire-tip absence was identified with 98%–99% accuracy over 690 images.
Takeaways & Limitations
VOIDD can automatically identify the navigated vessel and guidewire-navigation sequence, supporting semantic monitoring of successive PCI procedure steps.
Abstract
from arXiv · showhide
In this article, we present the work towards improving the overall workflow of the Percutaneous Coronary Interventions (PCI) procedures by capacitating the imaging instruments to precisely monitor the steps of the procedure. In the long term, such capabilities can be used to optimize the image acquisition to reduce the amount of dose or contrast media employed during the procedure. We present the automatic VOIDD algorithm to detect the vessel of intervention which is going to be treated during the procedure by combining information from the vessel image with contrast agent injection and images acquired during guidewire tip navigation. Due to the robust guidewire tip segmentation method, this algorithm is also able to automatically detect the sequence corresponding to guidewire navigation. We present an evaluation methodology which characterizes the correctness of the guide wire tip detection and correct identification of the vessel navigated during the procedure. On a dataset of 2213 images from 8 sequences of 4 patients, VOIDD identifies vessel-of-intervention with accuracy in the range of 88% or above and absence of tip with accuracy in range of 98% or above depending on the test case.
1 Introduction
PCI uses X-ray guidance and contrast injection, exposing patients and operators to ionizing radiation and limiting contrast tolerance. VOIDD combines contrast-enhanced reference images with guidewire-navigation fluoroscopy to identify the vessel being treated and the navigation period.
- PCI treats coronary artery stenosis with stents under X-ray guidance and contrast-agent injection.
- Contrast-agent tolerance is limited, while ionizing radiation affects both patients and operators.
- VOIDD combines contrast-enhanced reference images depicting vasculature with fluoroscopic navigation images acquired while the guidewire is advanced.
- The method automatically detects the vessel-of-intervention and identifies the guidewire-navigation period from subsequent fluoroscopic images.
- VOIDD addresses prior workflow limitations by matching guidewire-related fluoroscopy with vessel images, whereas related approaches may require operator-selected images and landmarks or omit motion compensation.
2 Vessel-of-intervention dynamic detection (VOIDD) algorithm
VOIDD tracks guidewire-tip and vessel candidates across navigation frames to select the vessel-of-intervention. It combines segmentation, vessel-graph extraction, curve matching, and temporally coherent track assignment.
- 2.1 General tracking framework: VOIDD detects guidewire tips and the vessel-of-intervention by grouping feature correspondences into tracks and selecting the most relevant track.Feature pairs link a navigation-sequence tip candidate with a reference-sequence vessel candidate.
- 2.1 General tracking framework: The algorithm ranks feature pairs by matching score, assigns them to existing tracks using TAD, and starts new tracks when the distance exceeds its threshold.Tracks contain at most one feature pair per navigation image.
- 2.2 Feature pairs extraction: Guidewire-tip candidates are extracted from min-tree components using elongation, with an upper bound suppressing elongated distractors such as pacing leads and filled catheters.The method exploits the guidewire tip’s thin, elongated appearance but does not rely on elongation thresholding alone.
- 2.2 Feature pairs extraction: Reference images are converted into vessel graphs from enhanced centerlines, with bifurcations and projected vessel crossings represented as graph nodes.A graph is computed for each reference frame across the cardiac cycle.
- 2.2 Feature pairs extraction: VOI candidates are generated by pairing tip centerlines with admissible vessel-graph paths using ECG information, curve matching, and residual Fréchet-distance similarity.Candidate paths are restricted near the tip candidate’s extremities to limit search complexity.
- 2.1 General tracking framework: TAD combines tip displacement, vessel-endpoint distance, and graph distance to preserve spatial and temporal coherence across tracks.The three distances are transformed into a common range using exponential functions parameterized by guidewire-tip length.
3 Results
VOIDD was evaluated for vessel-of-intervention and guidewire-tip navigation-sequence detection using annotated fluoroscopic sequences from four patients.
- A detection was correct when the tip-to-ground-truth vessel distance, measured by TRE, was below 0.5 mm.Missed detections were reported when the tip was present but the algorithm produced no detection; blurred tips and low contrast could cause misses.
- 88%–92% accuracy was achieved for locating the guidewire tip in the vessel-of-intervention across 1,513 navigation images from four patients.Sequences A1–D1 assessed vessel detection during guidewire navigation.
- 98%–99% accuracy was achieved for identifying sequences without a visible guidewire tip across 690 images.Sequences A2–D2 evaluated navigation-sequence identification when the tip was absent.
- The implementation required an average of 0.33 seconds for tracking on an Intel Core i7 processor running at 2.80 GHz.
4 Conclusion and future work
The paper presents VOIDD as a framework for determining the vessel-of-intervention and segmenting the guidewire tip during PCI fluoroscopy.
- VOIDD achieved a vessel-of-intervention detection success rate above 88% in the reported evaluation.
- Guidewire-tip tracking can identify guidewire-navigation phases and add semantic information for monitoring successive PCI procedure steps.
- The authors propose extending this approach to other interventional tools and collecting additional clinical cases in future work.