Real-time guidewire tracking and segmentation in intraoperative x-ray

TL;DR

This work tackles real-time guidewire segmentation and tracking in intraoperative X-ray fluoroscopy, where elongated deformable wires with low contrast are difficult to segment frame-by-frame. The authors propose a two-stage pipeline: a temporal-aware YOLOv5s detector with spatiotemporal refinement to localize candidate guidewires, followed by HessianNet, a lightweight segmentation network that uses a Hessian-based enhancement layer and a dual self-attention encoder–decoder to generate precise wire masks within detected boxes. Key contributions include (i) automatic thresholding and IOU-based box refinement to reduce false positives, (ii) the HessianNet architecture incorporating eigenvalue features $\lambda_1$ and $\lambda_2$ for robust segmentation, and (iii) real-time performance of approximately $35$ FPS on a GPU. The results demonstrate improved accuracy and robustness against low-quality images, enabling clearer visualization for clinicians and better support for robot-assisted interventions.

Abstract

During endovascular interventions, physicians have to perform accurate and immediate operations based on the available real-time information, such as the shape and position of guidewires observed on the fluoroscopic images, haptic information and the patients' physiological signals. For this purpose, real-time and accurate guidewire segmentation and tracking can enhance the visualization of guidewires and provide visual feedback for physicians during the intervention as well as for robot-assisted interventions. Nevertheless, this task often comes with the challenge of elongated deformable structures that present themselves with low contrast in the noisy fluoroscopic image sequences. To address these issues, a two-stage deep learning framework for real-time guidewire segmentation and tracking is proposed. In the first stage, a Yolov5s detector is trained, using the original X-ray images as well as synthetic ones, which is employed to output the bounding boxes of possible target guidewires. More importantly, a refinement module based on spatiotemporal constraints is incorporated to robustly localize the guidewire and remove false detections. In the second stage, a novel and efficient network is proposed to segment the guidewire in each detected bounding box. The network contains two major modules, namely a hessian-based enhancement embedding module and a dual self-attention module. Quantitative and qualitative evaluations on clinical intra-operative images demonstrate that the proposed approach significantly outperforms our baselines as well as the current state of the art and, in comparison, shows higher robustness to low quality images.

Figures (6)

• Figure 1: Our framework consists of two stages. First, the guidewire is detected in the current image frame at time step $T$ by including temporal information and the detection of the previous frame at time step $T-1$. Second, the guidewire is segmented in the predicted bounding box by our proposed HessianNet, which, in addition to the image information, utilizes Hessian-based features for improved guidewire segmentation.
• Figure 2: The architecture of the proposed Hessian Layer, which outputs multi-scale eigenvalue features $\lambda_{1}$ and enhances the guidewire in comparison to the original 2D images.
• Figure 3: Exemplary synthetic images and elastic deformation. A synthetic image (c) is generated based on source images (a) and (b). A elastic deformed image (e) and a boundary fixed elastic deformed image (f) is generated based on source image (d).
• Figure 4: Qualitative results of our and the state-of-the-art methods. Failure cases are highlighted with red arrows.
• Figure 5: Robustness of our and the baseline methods to variations in image brightness and contrast.