Encrypted login | home

Program Information

Megavoltage Iterative CT Reconstruction Utilizing L0-Penalty Based Total Count Variation Regularization

D Nguyen

D Nguyen1*, Q Lyu1 , D O'Connor1 , H Gao2 , X Qi1 , K Sheng1 , (1) Department of Radiation Oncology, University of California Los Angeles, Los Angeles, California, 90095 (2) Department of Radiation Oncology,Duke University Medical Center, Durham, North Carolina, 27710


SU-K-201-17 (Sunday, July 30, 2017) 4:00 PM - 6:00 PM Room: 201

Purpose: We present a novel iterative MVCT reconstruction method utilizing an isotropic L0-based total count variation (L0-TV) regularization for improving structure/ROI delineation.

Methods: We formulate the iterative CT reconstruction problem with an L2-norm fidelity term and an isotropic L0-TV regularization to enforce piecewise-constant regions. The non-convex formulation is efficiently solved to a suitable local minima with a special variant of the Primal Dual Hybrid Gradient (PDHG) algorithm, introduced in 2015 by Mollenhoff et al. This method was compared to filtered backprojection (FBP) and isotropic L1-TV iterative reconstruction using TomoTherapy MVCT data for an image quality phantom, head-and-neck (H&N) patient, and prostate (PST) patient. Reconstruction using 100% and 25% of the projection data were tested.

Results: For patient MVCTs, both TV methods shows improved noise reduction without noticeable loss in the anatomical details. The L0-TV produces images that have the best defined edges between different regions. For the 100% projection phantom data, the line pair profiles for line pairs 6 (0.3 lp/mm) and 7 (0.4 lp/mm) showed superior image resolution using the L0-TV method. The L1-TV performed similarly to FBP for line pair 6 and was inferior for line pair 7. For the 25% data, L0-TV managed to retain the same conspicuity as FBP for line pair 6. For line pair 7, L0-TV was able to resolve the pattern while L1-TV failed to do so. For contrast-to-noise ratio (CNR), the L0-TV method was superior to the other methods for the 100% projection data, and had similar CNR to the L1-TV method for the 25% data. FBP had the worst CNR in all cases.

Conclusion: L0-TV regularization in iterative MVCT reconstruction allows for higher CNR than FBP and better resolution retention than L1 regularization. This method is applicable to low dose kVCT. The piecewise-constant regions can improve delineation of low contrast structures/ROIs.

Funding Support, Disclosures, and Conflict of Interest: DOE DE-SC0017057 NIH R44CA183390 NIH R01CA188300 NIH R43CA183390 NIH U19AI067769

Contact Email: