Program Information
Spectral CT Reconstruction Via Patch-Based Low-Rank and Sparse Matrix Decomposition
S Niu1*, J Ma2 , J Wang3 , (1) ,,,(2) Southern Medical University, Guangzhou, Guangdong, (3) UT Southwestern Medical Center, Dallas, TX
Presentations
WE-DE-605-8 (Wednesday, August 2, 2017) 10:15 AM - 12:15 PM Room: 605
Purpose: Spectral CT can provide high energy resolution images using narrow energy bins. However, because the number of photons available in a narrow energy bin is small, and noise within each energy bin increases. In this study, we propose an iterative reconstruction method to effectively reduce the noise in spectral CT via patch-based low-rank and sparse matrix decomposition (PLSMD).
Methods: The PLSMD is used to exploit self-similarities in spectral CT images. Patches that share the same positon are collected and decomposed into a low-rank matrix and a sparse matrix. The low-rank matrix represents the stationary background, while the sparse matrix represents the rest of distinct intensity features. The resulting spectral CT reconstruction problem is efficiently solved using an alternating minimization algorithm. The performance of the proposed method is evaluated on a simulated PCD data. The quality of reconstructed spectral CT images is evaluated by comparing with those from filtered back-projection (FBP) method and robust principle component analysis (RPCA) method.
Results: The PLSMD method outperforms the RPCA method based on both visual inspection and quantitative measures. In the simulation study, the average local relative root mean square error (RRMSE) is reduced from 5.3% in RPCA to 3.1% in PLSMD, and the average structural similarity (SSIM) value increases from 0.78 in RPCA to 0.85 in PLSMD.
Conclusion: The PLSMD method can effectively suppress noise within narrow energy bin image while preserving spatial resolution. The PLSMD method improves image reconstruction accuracy of spectral CT as compared to RPCA method.
Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by the Cancer Prevention and Research Institute of Texas (RP130109),American Cancer Society (RSG-13-326-01-CCE), US National Institutes of Health (R01 EB020366), National Natural Science Foundation of China (81371544), Natural Science Foundation of Jiangxi Province (20161BAB212055), and Science and Technology Program of Jiangxi Education Committee (GJJ150994).
Contact Email: