Information in magnetic resonance images: evaluation of signal, noise and contrast

Abstract

The assessment of diagnostic image quality for MRI is considered. The assessment of three key image quality determinants is addressed: signal, noise and contrast. There is a distinction between random noise evaluation, for the calculation of the SNR, and structured noise evaluation for the assessment of image artefacts. Specific methods used are correlation techniques and the Wiener spectrum. Contrast is assessed by comparison of experimental data and theoretical predictions. For each assessment, the theory and method of the evaluation strategy are discussed. The discussion is illustrated with analysis results from commercial MR systems. The choice of analysis method and the subsequent derivation of quality indices are shown to be critical in respect of robustness and accuracy.

Thermoacoustic imaging over large field of view for three‐dimensional breast tumor localization: A phantom study

“…We will employ a transducer with a smaller focal spot size and a 3D reconstruction algorithm to further improve the Z-axis resolution. 30 Limited FOV has been a long-standing hindrance to the clinical application of breast cancer detection and localization in TAI. We have presented a unique USMP induced TAI system, which achieves high imaging depth, large FOV, and very low microwave radiation exposure.…”

“…5(c) and the SNR of the images can reach 14.5 dB. [29][30][31] The SNR of image was approximately calculated by the ratio of the maximum value of local variance of image to the minimum variance of the image. The experimental results showed that the 3D USMP TAI image could be used for tumor localization over a large FOV, again reinforcing the suitability of USMP TAI as an effective clinical diagnosis tool.…”

Direct comparison of myocardial perfusion cardiovascular magnetic resonance sequences with parallel acquisition

“…The high-pass spatial filter had a calibrated scale factor providing accurate noise measurement, which compensated for the change in noise level due to filtering. This method which used a high-pass spatial filter and temporal standard deviation is a generalization of the method using simpler spatial differentiation to provide improved suppression of the blood pool signal and extraneous artifacts (27). The method was validated against a volunteer study that used prescan noisebased SNR scaled reconstruction (26).…”

On the Expectation-Maximization algorithm for Rice-Rayleigh mixtures with application to noise parameter estimation in magnitude MR datasets

“…In the latter case, Sijbers et al (1998) provided a method that is robust to structural image errors and artifacts (Wilde et al, 1997), but requires data on k -space in addition to magnitude images. The single image techniques, on the other hand, attempt to automatically extract (Sijbers et al, 2007) the background by thresholding a histogram of the magnitude image data, and then estimate σ using maximum likelihood (ML) estimation on these Rayleigh distributed voxels (1).…”