Physics-Driven Passive Marker Tracking for MR-Guided Endovascular Interventions

Motivation: Endovascular interventions are compromised by the lack of interactive slice tracking during procedures. Goal(s): Using a physic-based marker model to establish an automated passive tracking method. Approach: Synthetic training data was generated and used to train a neural network for the detection of susceptibility markers. The dynamic tracking performance was evaluated in a vascular phantom. Results: The model can accurately simulate various passive marker appearances across arbitrary orientations and slice alignments, and facilitate AI-based passive tracking during MR-guided catheter interventions. Impact: Our physics-based approach allows for generating large annotated training datasets for passive marker detection, enabling robust tracking applications based on AI. This approach can improve the robustness during automated tracking of guidewires in MR-guided endovascular interventions.