Michael C Steckner1. 1. Toshiba Medical Research Institute USA, Inc., Mayfield Village, Ohio 44143, USA. msteckner@tmriusa.com
Abstract
PURPOSE: A simple extension to the NEMA MS-1 "difference of neighboring pixels" SNR method is presented, which can accurately determine the noise level within a signal region over a wide range of noise levels, image nonuniformities, and artifact levels, as demonstrated by simple simulations and experimental phantom images. METHODS: The new method computes difference of neighboring pixels in the read, phase, and diagonal directions. The variance of these three sets of pixel differences appear to contain the simple sum of the underlying variance of noise and any additional component unique to the read and phase directions, respectively, while the diagonal set of pixel differences contains all three components. By solving a set of three equations with three unknowns, it is possible to extract the components and isolate the desired noise variance term. RESULTS: Simulations show that the technique produces good results even when various artifact mechanisms present singly or jointly. Experimental results also demonstrate the technique works well but, depending on the severity of the artifact, cannot be guaranteed to always produce accurate results. CONCLUSIONS: Simulations and experimental results show the method to be accurate and robust. This method is applicable to multichannel receiver images, but not parallel reconstructed images.
PURPOSE: A simple extension to the NEMA MS-1 "difference of neighboring pixels" SNR method is presented, which can accurately determine the noise level within a signal region over a wide range of noise levels, image nonuniformities, and artifact levels, as demonstrated by simple simulations and experimental phantom images. METHODS: The new method computes difference of neighboring pixels in the read, phase, and diagonal directions. The variance of these three sets of pixel differences appear to contain the simple sum of the underlying variance of noise and any additional component unique to the read and phase directions, respectively, while the diagonal set of pixel differences contains all three components. By solving a set of three equations with three unknowns, it is possible to extract the components and isolate the desired noise variance term. RESULTS: Simulations show that the technique produces good results even when various artifact mechanisms present singly or jointly. Experimental results also demonstrate the technique works well but, depending on the severity of the artifact, cannot be guaranteed to always produce accurate results. CONCLUSIONS: Simulations and experimental results show the method to be accurate and robust. This method is applicable to multichannel receiver images, but not parallel reconstructed images.