Cardio-Respiratory Motion Estimation and Coronary Artery Segmentation for Image-Guided Percutaneous Coronary Intervention

Image guidance during percutaneous coronary interventions (PCI) can help minimize radiation exposure and contrast use while ensuring procedural safety and efficacy. To support this, this work proposes a framework that leverages a patient-specific cardio-respiratory motion model, optimized intra-procedurally, to enable real-time vessel tracking. The approach is built on: (i) a population-derived motion model capturing cardiac and respiratory dynamics, and (ii) an automated coronary artery segmentation pipeline for both 3D computed tomography angiography (CTA) and 2D x-ray angiography (XA). The motion model integrates cardiac phase and respiratory surrogates, including cycle phase and inhalation/exhalation ratio. To enable training and validation, paired 3D+t CTA and 2D+t XA sequences are synthetically generated using the proposed motion model. Coronary artery segmentation is performed using a dual-convolution-transformer U-Net. The approach was evaluated by comparing the segmented left ventricle across simulated and ground-truth 4D cardiac Magnetic Resonance Angiography datasets, demonstrating volume consistency within the 95% confidence interval. Segmentation achieved high Dice similarity scores: 0.86 ± 0.02 (CTA), 0.98 ± 0.01 (simulated XA), and 0.78 ± 0.01 (real XA). These results validate the accuracy of the synthetic motion simulation and segmentation pipeline. Future steps involve tracking of vessels by estimating patient-specific cardio-respiratory motion by using the proposed population-derived motion and segmented coronary arteries.