MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Stochastic Backprojection for Accelerated Model-Based Iterative 3D Image Reconstruction

A Sisniega1*, J Stayman1, S Capostagno1, C Weiss1, T Ehtiati2, J Siewerdsen1, (1) Johns Hopkins University, Balitmore, MD, (2) Siemens Healthineers, Forchheim, Germany

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: Model-based iterative reconstruction (MBIR) for cone-beam CT (CBCT) yields better noise-resolution tradeoffs and can accommodate noncircular source-detector trajectories. However, MBIR requires accurate forward and backprojection models, such as separable footprints (SF), that carry significant computational burden. Simple (mismatched) models offer reduced runtime, but inconsistencies in sampling between forward and backprojection operations causes sampling artifacts, especially conspicuous when voxel size is not matched to the detector pixel size. We report a sampling approach (“stochastic backprojection”) that overcomes such inconsistencies, helps to minimize runtime, and supports noncircular orbits.

Methods: The stochastic backprojection consists of a modification of the Peters backprojection operator by introducing a random perturbation of the ray position within each voxel, computed independently for every ray traced in the reconstruction. The approach was tested in CBCT on a robotic C-arm (Artis Zeego, Siemens) (496 projections, 100 kV, and 700 mAs) for an abdomen phantom with realistic anatomy (Kyoto Kagaku). Penalized weighted least squares (PWLS) reconstructions were obtained with matched SF operators and with the Siddon forward projector coupled to: i) Peters; and, ii) stochastic backprojector. Volumes of 342x342x240 mm were reconstructed with voxel size ranging from 0.4 mm (matched to detector pixel size) to 1.2 mm. Performance was assessed in terms of root mean squared error (RMSE) with PWLS using SF.

Results: Conventional Peters backprojector resulted in 5.9x reduced runtime compared to SF, but resulted in conspicuous sampling artifacts with RMSE ranging 4x10?³ mm?¹ and 1.8x10?² mm?¹ for 0.6 mm and 1.2 mm voxel size, respectively. The stochastic backprojector achieved 3.3x reduced runtime, with significant reduction of sampling artifacts for all voxel sizes (maximum RMSE of 4.0x10?4 mm?¹, for 1.2 mm voxels).

Conclusion: The stochastic backprojection method permits simplified, mismatched forward- and back-projectors in MBIR without the image quality degradation suffered with conventional, deterministic, backprojectors.

Funding Support, Disclosures, and Conflict of Interest: This research was supported by academic-industry partnership with Siemens Healthineers (AX Division, Forcheim, Germany).

Keywords

Cone-beam CT, Reconstruction, Image Artifacts

Taxonomy

IM- Cone Beam CT: Image Reconstruction

Contact Email