Grace Jianan Gang, Presenter: Nothing to Disclose

Joseph Webster Stayman PhD, Abstract Co-Author: Research Grant, Varian Medical Systems, Inc

Sarah Ouadah, Abstract Co-Author: Nothing to Disclose

Tina Ehtiati PhD, Abstract Co-Author: Employee, Siemens AG

Jeffrey H. Siewerdsen PhD, Abstract Co-Author: Research Grant, Siemens AG Consultant, Siemens AG Research Grant, Carestream Health, Inc Royalties, Elekta AB

New interventional cone-beam CT systems offer considerable freedom in source-detector positioning in addition to prior knowledge of the patient anatomy and intended imaging task. Such systems offer new opportunities to improve image quality and reduce dose. This work reports an optimization framework that identifies patient- and task-specific imaging protocols that maximize task performance and thereby reduce dose.

We employ a mathematical model of the imaging chain to compute task-specific detectability index (d') for given patient anatomy as a function of the acquisition protocol (including source, detector, and orbit) and reconstruction method (algorithm and parameters therein). Using detectability index as the objective function, an optimization framework was established to identify protocols best suited for a given task. Tube current, parameterized by a scalar , was optimized via exhaustive search keeping total dose within a predefined constraint; using the optimal , reconstruction parameter [e.g., cutoff frequency (f0) in filtered-backprojection (FBP)] was optimized exhaustively for each view. The source-detector trajectory was optimized using a greedy algorithm that successively finds the next best view (angle/obliquity) for maximum improvement in d'. Task-driven current + kernel modulation was applied to a line pair detection task in an elliptical phantom. Trajectory optimization was performed for an intracranial hemorrhage detection task and compared with a circular orbit. Experiments were performed on an experimental CBCT bench.

At the same total mAs, the line pair pattern was conspicuous for task-driven (current + kernel) modulation case but barely distinguishable in the unmodulated and AEC cases. Task-based trajectories designed for hemorrhage detection successfully avoided highly attenuating rays associated with the embolization coils and skull base, yielding reconstructions with major reduction of noise and artifacts.

Compared to conventional acquisition and reconstruction, a task-driven imaging approach tailored to specific patient anatomy and imaging tasks demonstrates the potential for improved image quality and reduced dose.

The task-driven framework synergizes advances in image quality modeling, advanced C-arm systems (e.g., Zeego), and model-based reconstruction to achieve improved image quality and reduced dose.

Gang, G, Stayman, J, Ouadah, S, Ehtiati, T, Siewerdsen, J, Task-Driven Image Acquisition and Reconstruction in Cone-Beam CT for Interventional Guidance.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14019747.html