Automated Passive Tracking for MR-guided Endovascular Interventions

Vascular and cardiac interventions often rely on X-ray fluoroscopy for real-time imaging guidance, which poses risks due to ionizing radiation exposure. MRI offers a safer alternative with superior soft tissue contrast and avoids using ionizing radiation, but its adoption for near real-time guidance so far has been limited by technical challenges. This paper presents a novel approach for real-time detection and tracking of an MR-visible guidewire with discrete markers during endovascular interventions, addressing key limitations. A series of experiments was designed to evaluate and optimize marker visibility in terms of contrast-and-noise for different sequences. A convolutional neural network (CNN) is trained to detect the marker artefacts using a physics-based marker model and evaluated experimentally in an aortic phantom. The physics-based model simulation is used to generate the dataset without the need for MRI imaging, or manually annotating dataset for training the CNN model. The output from the CNN model is integrated into a custom software interface that links the tracked markers to automatic MRI slice geometry alignment, enabling near real-time guidewire tracking. Experimental results evaluate the effectiveness of the proposed approach in detecting the guidewire under varying imaging conditions, showing promise in improving the safety, efficiency, and accuracy of MR-guided interventions.