Assistant Professor Clemson University, South Carolina, United States
Introduction: : Intravascular Ultrasound (IVUS) catheters are indispensable tools for minimally invasive cardiovascular interventions, including percutaneous coronary interventions (PCI) and peripheral artery disease (PAD) treatments. These devices provide cross-sectional ultrasound images of blood vessel lumens and walls, offering critical insights into plaque morphology, lesion severity, and stent apposition. The ultrasound image modality can also be used for navigation within complex vascular anatomies. However, accurate three-dimensional localization of the catheter tip within the vessel lumen remains a significant challenge, particularly in vessels with numerous side branches. This difficulty often necessitates the use of supplementary imaging modalities, such as fluoroscopy, to guide catheter placement and ensure procedural safety and efficacy. However, the reliance on additional imaging inherently exposes both patients and medical staff to undesirable radiation. To mitigate this concern, we developed an image analysis tool designed to enhance IVUS catheter localization without the need for additional radiation exposure. Specifically, this project focuses on the image analysis of two-dimensional ultrasound images acquired directly by the IVUS probe, enabling the localization of potential side branches. This analysis will allow us to map the acquired 2D images onto a pre-existing three-dimensional vessel model to allow for precise localization. This 3D model is reconstructed from a pre-operative Computed Tomography (CT) scan, providing a comprehensive anatomical reference. This advancement promises to improve procedural accuracy, reduce radiation burden, and ultimately enhance patient outcomes in a wide range of endovascular procedures.
Materials and
Methods: : We used image processing and pixel analysis to detect side branches in an IVUS image. After binarizing the image and removing small pixel groups, we established region boundary lines to determine if there is a hole in the circular outline of the vessel that could be a side branch. Using IVUS imaging conventions, we identified the catheter as the center of the image. Radiating from this center with a number of evenly spaced spokes, we defined the edges of the main vessel. We then used groups of three of these edge points to calculate many potential centerpoints of the main vessel, and excluded outlier candidates. Due to the nature of IVUS, the vessel wall outline has a jagged edge which can lead to calculated centers outside of the predetermined vessel edge. These outliers were identified and removed before averaging those remaining to establish a center to the main vessel. Using the vessel edges and center, we calculated an average radius and constructed a circle around the center. This circle defined a path to determine the location of the side branch. Each pixel location along the path was checked to see if it was part of the vessel wall. Once a gap was identified, we recorded the orientation of the branch, which will allow us to map the analyzed image onto the 3D vessel model. We created such a 3D vessel model by segmenting a CT scan to isolate the aorta from the arch to the common iliac arteries.
Results, Conclusions, and Discussions:: Our algorithm to identify side branches in IVUS images was created based on a starter image. After the algorithm was complete, additional images were acquired from the internet to test and improve the robustness of the system. These images were from different sources, so the images varied greatly in image quality, image modality, and type of vessel seen. Out of the ten additional images tested, nine were successfully processed with our algorithm. Of the images that were successfully processed, five did not have a side branch, two did have a side branch, and two possibly contain a guide wire shadow. In all of these cases, the system found the main vessel edge and a reasonable center to the vessel, which is different from the center of the image where the catheter is located. From these points, the orientation of the branch was successfully calculated in relation to the center of the vessel. Due to the limited availability of clear side branch IVUS images and corresponding 3D imaging publicly available, we could not yet test our algorithm completely. Two of the images we tested had what appeared to be a guide wire shadow, but the algorithm saw them as side branches. With more images to test, we can alter our algorithm to distinguish between side branches and guide wire shadows. To mitigate this limitation, we plan to simulate ultrasound images based on CT segmentation data. By using the 3D model data and the calculated centerline, we can simulate the catheter moving through the vessel and calculate what an ultrasound image would look like. This would allow us to create more images for further algorithm testing, including many more branched images than is freely available online. Creating these images will also allow us to advance to the next stage in the research process, mapping the 2D IVUS images to the 3D model, which requires corresponding 2D and 3D data. Once our algorithms reliably work in simulation, we plan to test them on a physical IVUS system.
