Joseph Webster Stayman PhD, Presenter: Research grant, Varian Medical Systems, Inc

Wojciech Zbijewski PhD, Abstract Co-Author: Research Grant, Carestream Health, Inc

Yoshito Otake, Abstract Co-Author: Nothing to Disclose

Yifu Ding, Abstract Co-Author: Nothing to Disclose

Hao Dang, Abstract Co-Author: Nothing to Disclose

Sebastian Schafer, Abstract Co-Author: Nothing to Disclose

Gary Gallia, Abstract Co-Author: Nothing to Disclose

Marc Sussman, Abstract Co-Author: Nothing to Disclose

Jeffrey H. Siewerdsen PhD, Abstract Co-Author: Advisory Board, Siemens AG Advisory Board, Carestream Health, Inc Research Grant, Siemens AG Research Grant, Carestream Health, Inc

Statistical reconstruction methods now increasingly available for diagnostic CT have yet to be fully realized in interventional imaging. Not only does the acquisition physics differ between diagnostic and interventional systems, but the images serve very different purposes (e.g., localization vs detection) and strong prior information is almost always available (e.g., preop images). This work applies novel model-based reconstruction methods leveraging system-specific physics and patient-specific prior knowledge in intraoperative cone-beam CT (CBCT) to bring the advantages of dose reduction and image quality to the interventional suite.

A general framework for advanced reconstruction in C-arm CBCT has been developed that includes a high-fidelity noise model and allows incorporation of patient-specific prior knowledge. These methods are amenable to a spectrum of interventional applications. Considered here are i) intraoperative detection of intracranial hemorrhage (ICH), and ii) needle localization in image-guided biopsy/ablation. The first stresses the limits of CBCT contrast resolution and exercises the power of high-fidelity statistical reconstruction, while the second utilizes prior information (an initial CBCT) for fast, sparse-projection 3D imaging of needle placement. Experiments were conducted using a prototype CBCT C-arm and a dedicated x-ray testbench.

For ICH detection, the model-based approach demonstrated increased low-contrast detectability (at matched spatial resolution) at equivalent dose (2.8 mGy) compared to traditional Feldkamp reconstruction, providing ~1 cm3 bleed visibility across the full range of relevant contrast (-30 to +70 HU). For needle localization in simulated biopsy procedures, incorporation of an initial (fully sampled) CBCT in 18x sparse acquisitions demonstrated dramatic dose and data reduction while preserving image quality and geometric accuracy of localization.

The ability of advanced reconstruction methods to model the system-specific physics and to incorporate patient-specific knowledge (each typically ignored in interventional imaging) permits improved imaging performance and dose utilization, and highly undersampled data without loss in image quality.

Advanced reconstruction techniques that emphasize statistical models and patient-specific priors will enable higher-quality, lower-dose imaging in image-guided interventions.

Stayman, J, Zbijewski, W, Otake, Y, Ding, Y, Dang, H, Schafer, S, Gallia, G, Sussman, M, Siewerdsen, J, Advanced 3D Statistical Reconstruction and Incorporation of Prior Knowledge for High-Quality, Low-Dose C-Arm Cone-Beam CT in Image-Guided Interventions.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12034285.html