Towards Model-Enhanced Real-Time Ultrasound Guided Cardiac Interventions

Image-guidance is a key component to the development of new minimally invasive off-pump cardiac procedures, and can be provided by combining high resolution pre-operative models with real-time intra-operative imaging. We present three real-time registration approaches for model-enhanced guidance, exploring different regions of the heart and both surface-based and intensity-based registration. The surface-based registrations employ a continuous max flow graph cut for surface extraction, while the image-based registration relies on intelligent image sub sampling to reduce computational requirements. With multithreaded general purpose programming on graphics processor units (GPGPU), we demonstrate that these approaches can be implemented at sufficient speed and accuracy for intra-operative guidance.