Andreas Horlbeck: CNN-based Implant Detection and Ellipsoidal Bounding Box Parametrization from Dual View X-Ray Images
Mobile C-Arm imaging systems are frequently used during surgery for image guidance and implant placement verification. Two dimensional images are routinely acquired prior to a CBCT 3D scan to align the system with the patient anatomy. The goal of this work is to develop a deep learning-based approach to detect and parametrize an ellipsoidal bounding box around metallic objects from two available projection images. This description of the metal implants is used for scene understanding and additional downstream adaptations to the imaging workflow to enhance image quality and reduce dose.