Program Information
Pixel-Wise Calculation of Noise Statistics On Iterative CT Reconstruction From a Single Scan
T Wang*, L Zhu , Georgia Institute of Technology, Atlanta, GA
Presentations
SU-C-206-5 (Sunday, July 31, 2016) 1:00 PM - 1:55 PM Room: 206
Purpose: As iterative CT reconstruction continues to advance, the spatial distributions of noise standard deviation (STD) and accurate noise power spectrum (NPS) on reconstructed CT images become more important for method evaluation ans algorithm parameters settings. Using data of a single scan, we propose a practical method for pixel-wise noise statistics calculation on iterative CT reconstruction image, which enables accurate calculation of noise STD for each pixel and NPS without stationary noise assumption.
Methods: We first iteratively reconstruct a CT image using existing reconstruction algorithm. Noise perturbation is then simulated on the raw projection data based on the estimated noise statistics, and the iterative reconstruction continues to generate a different CT image. The above process is repeated on the same projection data but with different noise perturbations, until a sufficient number (32 in our implementations) of different noisy reconstruction images are generated. Pixel-wise calculations of noise STD and NPS are finally performed on the entire stack of noisy images.
Results: We have evaluated our method via simulation studies on a clinical head-and-neck CT image, using two iterative reconstruction algorithms: penalized weighted least-square (PWLS) and total-variation (TV) regularization. 128 sets of simulated noisy sinogram are used to generate ground-truths of noise STD and NPS. Using only one set noisy sinogram, the proposed method accurately calculates noise STD map, with a correlation of 94% and 95% for PWLS and TV methods, respectively, compared with the ground-truth. The relative root-mean-square error of 1D-NPS for proposed method are 1.5% and 3.4% for PWLS and TV methods, respectively, while those for conventional method (averaging NPS of sub-ROIs in ROI of one image) are 9.0% and 12.9%.
Conclusion: The proposed method accurately calculates noise STD for each pixel and NPS without stationary noise assumption on iteratively reconstructed CT image, with no requirements of repeated CT scans.
Funding Support, Disclosures, and Conflict of Interest: Research reported in this abstract was supported by the National Institute Of Biomedical Imaging And Bioengineering of the National Institutes of Health under Award Number R21EB019597.
Contact Email: