Measurement of signal‐to‐noise ratios in MR images: Influence of multichannel coils, parallel imaging, and reconstruction filters

Abstract

Purpose: To evaluate the validity of different approaches to determine the signal-to-noise ratio (SNR) in MRI experiments with multi-element surface coils, parallel imaging, and different reconstruction filters. Materials and Methods:Four different approaches of SNR calculation were compared in phantom measurements and in vivo based on: 1) the pixel-by-pixel standard deviation (SD) in multiple repeated acquisitions; 2) the signal statistics in a difference image; and 3) and 4) the statistics in two separate regions of a single image employing either the mean value or the SD of background noise. Different receiver coil systems (with one and eight channels), acquisitions with and without parallel imaging, and five different reconstruction filters were compared. Results:Averaged over all phantom measurements, the deviations from the reference value provided by the multiple-acquisitions method are 2.7% (SD 1.6%) for the difference method, 37.7% (25.9%) for the evaluation of the mean value of background noise, and 34.0% (38.1%) for the evaluation of the SD of background noise. Conclusion:The conventionally determined SNR based on separate signal and noise regions in a single image will in general not agree with the true SNR measured in images after the application of certain reconstruction filters, multichannel reconstruction, or parallel imaging. THE SIGNAL-TO-NOISE RATIO (SNR) is an important quantity used to describe the performance of a magnetic resonance imaging (MRI) system, and is frequently used for image evaluation, measurement of contrast enhancement, pulse sequence and radiofrequency (RF) coil comparison, and quality assurance. Several methods to determine the SNR of MR images have been described. The most commonly used technique is based on the signal statistics in two separate regions of interest (ROIs) from a single image: one in the tissue of interest to determine the signal intensity, and one in the image background to measure the noise intensity (1,2). There are two important preconditions for SNR measurements based on this "two-region" approach: a spatially homogeneous distribution of noise over the whole image is required, and the statistical intensity distribution of the noise should be known so that the noise properties measured in a background area can be used to deduce the noise distribution overlaying the anatomic structures in the foreground. These assumptions have been valid for many MR images in past years, particularly for those acquired by spin-warp imaging (image reconstruction by 2D or 3D Fourier transform) with a standard single-channel volume quadrature coil followed by a magnitude reconstruction.However, the use of newly developed phased-array surface coil systems and new reconstruction techniques, such as parallel imaging (3,4), can influence both the statistical and the spatial distribution of noise. For instance, the noise distribution in parallel imaging is described by the spatially varying geometry factor (g-factor) and depends on parameters such as the coil geometry, phase-enc...

Quantification of pulmonary blood flow (PBF): Validation of perfusion MRI and nonlinear contrast agent (CA) dose correction with HO positron emission tomography (PET)

“…When using parallel imaging in low signal regions like lung parenchyma, calculation of SNR or contrast to noise ratio (CNR) is not reliable (36). Hence, signal-time curves were compared by peak relative signal intensities (PRSIs; ratio of peak SI during peak CA concentration and the average of eight baseline signal intensities) instead.…”

“…The amount of noise and hence its effect on the correction scheme for this study can only be roughly estimated: use of the GRAPPA parallel imaging technique renders the measurement of noise in an ROI outside the body unreliable (36). The estimation of the mean noise bias was 2.4 (cf.…”

Half‐fourier‐acquisition single‐shot turbo spin‐echo (HASTE) MRI of the lung at 3 Tesla using parallel imaging with 32‐receiver channel technology

“…The median rank for intrapul- monary anatomy, artifacts and the subjectively perceived SNR for all acceleration factors was calculated. The SNR of HASTE was analyzed in a water phantom for all acceleration factors using a difference method, as previously described (21). For this, the signal intensity in a circular region of interest was divided by the standard deviation of the difference signal of two baseline images in the same region.…”

Improving quality of arterial spin labeling MR imaging at 3 tesla with a 32‐channel coil and parallel imaging

“…The contrast-to-noise ratio (CNR) was computed as 20 log10((SGM-SWM)/s) where SGM and SWM are the measured signal values on perfusion-weighted images over GM and WM and s is the mean temporal SD on the ROI (16). The signal-to-noise-ratio (SNR) in dB was computed as 20 log10(S/s) where S is the measured signal value of control raw images or CBF over a ROI and s is the mean temporal SD on the ROI (5,9).…”