PROJECT SUMMARY / ABSTRACT While intraoperative C-arm cone-beam CT (CBCT) is used in a growing spectrum of minimally invasive image- guided procedures, imaging performance lags behind conventional diagnostic CT – often due to challenges particular to the interventional imaging suite. However, interventional imaging also has a number of distinct ad- vantages with prior studies and planning data providing information about patient anatomy, imaging targets, and potential imaging tasks. Coupling this with the increased flexibility of interventional systems to automatically position the source and detector in arbitrary geometries opens a unique opportunity to do something diagnostic CT cannot do. Specifically, one can use all of the patient-, task-, and procedure-specific information to drive these intervention systems in customized orbits wherein the imaging device, with the help of a mathematical model of imaging performance prediction, can prospectively drive a source-detector trajectory to acquire the most information rich views. Such a “smart” imaging system could automatically maximize imaging performance with better image quality, lower radiation doses, etc. without relying on the experience or expertise of a CT technician. We propose a framework that integrates prior anatomical knowledge and treatment plans, the capability for non-circular orbits, and a definition of the imaging task to drive custom task- and patient- specific source-detector trajectories for 3D interventional imaging. We seek to achieve this task-driven interventional imaging through the following specific aims: Aim 1: Develop foundations for task-driven inter- ventional imaging including a quantitative framework for task-based performance prediction, an optimizer that maximizes regional performance for user-defined tasks. Aim 2: Develop clinical systems for task-driven in- terventional imaging including methods to command two state-of-the-art clinical C-arms – a floor-mounted robotic C-arm and a mobile C-arm, Cios Spin – in optimized task-driven trajectories. Aim 3: Conduct clinical pilot studies for an initial evaluation of task-driven imaging in patients. A pilot study is planned for two clinical target applications: prostatic artery embolization and cervical spine fusion which are traditionally chal- lenged by difficult anatomy and interventional hardware that can severely degrade measurements in specific views. Safety and feasibility of the proposed task-driven imaging approach will be established and preliminary data on relative performance and observer preference test will be conducted to inform future clinical studies. Successful completion of these aims begins opens a new paradigm for task-driven interventional imaging that rigorously leverages patient-specific prior information to acquire and reconstruct data that is optimal for specific interventional imaging tasks.

PROJECT NARRATIVE 3D-capable C-arms continue to evolve with increased use across a range of interventional applications with increased flexibility and automation in their movements; however, these systems have yet to fully integrate the wealth of knowledge about the specific patient anatomy being imaged and the specific task being conducted with their advanced acquisition capabilities and the advent of sophisticated reconstruction methods. We propose a novel framework that breaks from traditional data acquisition models leveraging rigorous definitions of task in a patient- and data-dependent performance prediction framework to automatically drive state-of-the-art C-arms in optimized noncircular orbits; and through model-based iterative reconstruction produce superior image quality and dose utilization over traditional acquisition and processing approaches. While we target two particularly challenging imaging scenarios (namely, imaging of the prostate which is surrounded by the dense bone of the pelvis; and imaging of the cervical spine which is challenged both by fixation hardware and the anatomy of the shoulders), these approaches are generally applicable, seeking to acquire the best data for the imaging scenario and pursuing a new paradigm for imaging devices driven by patient-specific and task-specific knowledge.