Automated Generation of Synthetic X-Ray Datasets in Percutaneous Coronary Intervention (PCI) – Research Project by Mersad Shoaei Taklimi

Join us for the presentation of a master’s project on creating an automated pipeline for synthetic X-ray dataset generation in percutaneous coronary intervention (PCI). PCI is one of the most important and widely used minimally invasive interventions to treat coronary artery disease and it relies heavily on real-time X-ray fluoroscopy for navigation of devices inside the coronary vessels. However, training and evaluating the AI models for this purpose is diminished by the scarcity of large and well-annotated datasets.

This project uses Blender-based 3D modeling of PCI devices like stents, guidewires, balloons, and other devices, together with physics-informed forward projection using DeepDRR to produce realistic X-ray images followed by a precise segmentation mask of all devices. Our work presents a fully automated pipeline from device simulation and voxelization to projection onto CT-derived anatomical backgrounds to achieve high degree of visual similarity to clinically acquired fluoroscopy data.