Premium
SU‐E‐I‐70: Quantification of Noise Reduction Dependency On Spatial Granularity With Adaptive Statistical Iterative Reconstruction
Author(s) -
Zhou Y,
Scott A
Publication year - 2014
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 - 10.1118/1.4888020
Subject(s) - imaging phantom , iterative reconstruction , noise reduction , noise (video) , contrast to noise ratio , image noise , mathematics , projection (relational algebra) , reduction (mathematics) , image resolution , nuclear medicine , physics , optics , geometry , image quality , algorithm , artificial intelligence , acoustics , computer science , image (mathematics) , medicine
Purpose: Adaptive statistical iterative reconstruction (ASIR) is often used for noise reduction. However, it may not be clear whether the noise reduction is uniform across different spatial scales. As this is important for lesion detection, we attempted to quantify the noise reduction dependency on spatial scales under different ASIR blending fractions. Methods: An abdomen phantom (CIRS TE‐07) simulating the medium sized patient was used. The phantom contains a cylindrical void of 4 cm diameter, filled with an insert containing six groups of cylindrical targets (1% and 2% contrast) of 1.2 mm to 7 mm. Helical scans were conducted using a GE 750HD with 120 kVp, 330 mAs and pitch of 1.375 (CTDI_vol = 18 mGy). The images were reconstructed to 5 mm thickness with filtered‐back projection (FBP) and with ASIR at blending fractions of 10% to 100%. The uniform sections of the contiguous slices were subtracted and the resulted regions were divided into matrices of cells matching the target size from 1.2 mm–7.0 mm. Standard deviations (sd) from the means of all cells were computed as a measure of noise at the target size. The noise reduction factor (NRF) was defined as the ratio of the sd from ASIR to the sd of FBP. The NRF was fitted to ASIR blending percentages (ASIRP) at different target sizes (d). Results: The NRF versus ASIRP was fitted to a linear relationship with an intercept of approximately 1.0 and a slope dependent on d (R square = 0.99). The slope versus d was fitted to a power relationship with a power index of – 0.80 and a proportion factor of −0.005 (R square = 0.978). Conclusion: The noise reduction with ASIR increases nonlinearly as the spatial scale (or target size) decreases. The result can be of help for ASIR use in target‐specific diagnoses.