Enhancing Fluoroscopy-Guided Interventions: a Neural Network to Predict Vessel Deformation without Contrast Agents

Abstract

Image-guided procedures have experienced a rapid in- crease in popularity in recent years. The advancements in medical imaging technology have led to a shift in medical images from being used primarily for diagnosis to being a critical tool in theragnostic and therapeutic procedures. This shift has resulted in the emergence of new fields such as interventional radiology (IR), thera- peutic endoscopy (TE), and minimally invasive image- guided surgery (IGS), with an increasing number of professionals adopting these techniques in their clinical practices due to improved outcomes [1]. One of the most widely used imaging methods in these procedures is X-ray-based imaging, including computed tomography (CT), 2D C-arm fluoroscopy, and cone-beam CT scans. These procedures typically require the use of contrast agents (CA) to visualize soft tissues with high definition and contrast. However, the use of CA presents several challenges, including the limited volumes that can be used and the toxicity of the agents when they are injected intravascularly [2]. The CA also follows the patient’s hemodynamics, leading to transient visualization and asynchronous image guidance. In this paper, we aim to address the technical issues related to contrasted X- Ray images in image-guided therapy. We propose a deep learning approach that will allow for the visualization of vessels during image-guided procedures without the need for contrast agents, making these procedures safer, and more efficient, while providing real-time guidance.