Premium
TU‐F‐CAMPUS‐I‐01: Statistical Iterative Reconstruction for Perfusion CT with a Prior‐Image Induced Hybrid Nonlocal Means Regularization
Author(s) -
Li B,
Lyu Q,
Ma J,
Wang J
Publication year - 2015
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4925796
Subject(s) - iterative reconstruction , regularization (linguistics) , imaging phantom , algorithm , mathematics , similarity (geometry) , computer science , image quality , nuclear medicine , artificial intelligence , image (mathematics) , medicine
Purpose: In CT perfusion imaging, an initial phase CT acquired with a high‐dose protocol can be used to improve the image quality of later phase CT acquired with a low‐dose protocol. For dynamic regions, signals in the later low‐dose CT may not be completely recovered if the initial CT heavily regularizes the iterative reconstruction process. To overcome the limitation of the conventional prior image induced penalty, we propose a hybrid nonlocal means (NLM) regularization for iterative reconstruction of perfusion CT. Methods: The hybrid penalty is constructed by combining the NLM of initial high‐dose CT in the stationary region and later low‐dose CT in the dynamic region. The stationary and dynamic regions are determined by the similarity between the initial high‐dose scan and later low‐dose scan, where the similarity is defined as Gaussian distance between patch‐window of the same pixel of the two scans. The similarity measure is then used to weight the influence of the initial high‐dose CT. For regions with high similarity (e.g., stationary region), initial high‐dose CT will play a dominant role in regularizing the solution. For regions with low similarity (e.g., dynamic region), the regularization will rely on low‐dose scan itself. This new hybrid NLM (hNLM) penalty is then incorporated into the penalized weighted least‐squares (PWLS) for perfusion CT reconstruction. Digital and anthropomorphic phantom studies were performed to evaluate the PWLS‐hNLM algorithm. Results: Both phantom studies show that the PWLS‐hNLM algorithm is superior to the conventional penalty term without considering the signal changes within dynamic region. In the dynamic region, the reconstruction error measured by root mean square error is reduced by 50% in PWLS‐hNLM reconstructed image. Conclusion: The PWLS‐hNLM algorithm can effectively use initial high‐dose CT to reconstruct low‐dose perfusion CT in the stationary region while avoiding its influence in the dynamic region.