Program Information
Improve the Resolution of Statistical Iterative CBCT Reconstruction by Considering Both Hardware Blur and Software Blur
Q Shi1 , L Liu1 , J Wang2 , S Tan1*, (1) Huazhong University of Science and Technology, Wuhan, Hubei, China (2) UT Southwestern Medical Center, Dallas, TX
Presentations
SU-I-GPD-I-12 (Sunday, July 30, 2017) 3:00 PM - 6:00 PM Room: Exhibit Hall
Purpose: To develop a new statistical iterative reconstruction (SIR) method that accounts for both hardware blur and software blur in CBCT.
Methods: Detector blur (hardware blur) in the projection data was taken into account in the SIR objective function, leading to a non-diagonal noise covariance matrix. The detector blur was approximated using Gaussian function and estimated from the projection data by placing a tungsten edge on the face of the flat-panel detector. The Hessian penalty can effectively suppress the staircase effect of the total variation (TV) penalty with an extra cost of blurring object edges (software blur). We combined TV and Hessian in a structure adaptive way to reduce this kind of blur due to the regularization. A generalized least-squares penalized weighted least-square (PWLS) criterion was designed to account for both detector blur and regularization blur. An effective algorithm was developed to minimize the objective function using a majorization-minimization (MM) approach. We evaluated the proposed method on a Compressed Sensing (CS) and a modified 3D shepp-logan phantom. The data were reconstructed with different regularizations using both uncorrelated model (not accounting for detector blur) and correlated model (accounting for detector blur). The full-width-at-half-maximum (FWHM) was calculated for comparison.
Results: Both blur considerations improved the resolution of the reconstruction results. For the CS phantom, images reconstructed using the uncorrelated model had lower spatial resolution than those reconstructed using the correlated model at the same noise level. For the shepp-logan phantom, images reconstructed using the combined penalty preserved edges like TV and preserved low-contrast objects like Hessian. The FWHM due to the combined penalty was similar to TV, and less than that of Hessian.
Conclusion: The proposed SIR considering both detector blur and regularization blur improved the spatial resolution compared to traditional SIR.
Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by National Natural Science Foundation of China (NNSFC), under Grant Nos. 61375018 and 61672253. J. Wang was supported in part by grants from the Cancer Prevention and Research Institute of Texas (RP130109 and RP110562-P2), the National Institute of Biomedical Imaging and Bioengineering (R01 EB020366).
Contact Email: