DOI: 10.1148/radiol.2433060996
(Radiology 2007;243:908-909.)
© RSNA, 2007
Measurement of Signal-to-Noise Ratio in MR Imaging with Sensitivity Encoding
Won-Jin Moon, MD, PhD
Department of Radiology, Konkuk University Hospital, 4-12, Hwayang-dong, Gwangjin-gu, Seoul 143-914, Korea
e-mail: mdmoonwj{at}naver.com
Editor:
I read with much interest the article by Dr Kuhl and colleagues (1) in the February 2005 issue of Radiology. The authors stated that signal-to-noise ratio (SNR) for a diffusion-weighted magnetic resonance (MR) sequence with parallel imaging was inferior to that without parallel imaging. They measured SNR as a quotient of the average signal intensity (SI) in a region of interest (ROI) within the brain and the standard deviation of the noise in an ROI outside the brain. But I have some concern about the noise characteristics of parallel imaging by using this conventional SNR measurement. As described by Pruessmann et al (2), the SNR of parallel imaging can be described as follows: SNR = SNR/(geometry factor x sensitivity encoding [SENSE] factor). The geometry factor represents noise amplification across the image induced by using an unwrapping algorithm for different coil elements. Of importance, the geometry factor is dependent on the spatial location within the range, the SENSE factor, and the geometry of the specific coil array. For these reasons, measurement of noise within in a region other than the region where SI is measured may lead to an erroneous estimation of local SNR.
Therefore, there are alternative methods for measurement of SNR with parallel imaging (3,4). SNR for MR imaging with SENSE can be measured through the repeated acquisition of multiple images with identical imaging parameters. Although this approach is robust and accurate, it can be time consuming and may be impractical (3,4). Another method for measuring SNR can be performed through the acquisition of two identical images. In this approach, an estimate of the mean SI is obtained from a small ROI from the sum of SI 1 and SI 2, and the standard deviation of the difference of SI 1 and SI 2 is obtained from the same ROI (4).
What is also intriguing is how Dr Kuhl and colleagues could obtain the lower SNR with parallel imaging (exactly revealed theoretical SNR loss with parallel imaging) by using conventional SNR measurement. All three vendors of the equipment used in the study implemented background noise elimination software in their diffusion-weighted imaging sequences. Therefore, when one measures the SNR of a background ROI with SENSE and without SENSE, one can get a higher SNR with SENSE imaging.
 |
References
|
|---|
- Kuhl CK, Gieseke J, von Falkenhausen M, et al. Sensitivity encoding for diffusion-weighted MR imaging at 3.0 T: intraindividual comparative study. Radiology 2005;234:517–526.[Abstract/Free Full Text]
- Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: sensitivity encoding for fast MRI. Magn Reson Med 1999;42:952–962.[CrossRef][Medline]
- Preibisch C, Pilatus U, Bunke J, Hoogenraad F, Zanella F, Lanfermann H. Functional MRI using sensitivity-encoded echo planar imaging (SENSE-EPI). Neuroimage 2003;19:412–421.[CrossRef][Medline]
- de Zwart JA, Ledden PJ, van Gelderen P, Bodurka J, Chu R, Duyn JH. Signal-to-noise ratio and parallel imaging performance of a 16-channel receive-only brain coil array at 3.0 Tesla. Magn Reson Med 2004;51:22–26.[CrossRef][Medline]
Response
Christiane K. Kuhl, MD,
Jürgen Gieseke, PhD, and
Hans H. Schild, MD
Department of Radiology, University of Bonn, Sigmund-Freud-Strasse 25, D-53105 Bonn, Germany
e-mail: kuhl{at}uni-bonn.de
Measuring SNR on MR images that were acquired with parallel imaging techniques has been challenging—for the very reasons correctly mentioned by Dr Moon. However, this is exactly why we offered three different approaches to rate the SNR (1): first the "conventional" way, that is, by referring the tissue SI to the background noise (with the inherent difficulty mentioned above); then by referring brain tissue SI to that of a "tissue" with persistently very low SI, that is, cerebrospinal fluid ("relative" SI); and last by simply assessing SNR visually. The fact that these multiple approaches were made acknowledges the well-known difficulties associated with SNR measurements on SENSE images. This fact is also explicitly explained in the materials and methods section of our article: "The rationale of using relative SI instead of the usual SNR was to provide a rough estimate of the overall SI of the image without the possibly misleading contribution of image noise."
Adequate techniques to quantify SNR on SENSE images were published only a year after our study had been submitted. However, since we had been aware of the problem, the "conventional" SNR data were indeed interpreted with reservation and were presented in our article as being possibly misleading. Accordingly, our conclusions match exactly with what Dr Moon suggests: "The nominal loss of SNR that occurred as a result of use of SENSE did not translate into reduced image quality. In fact, the images appeared to offer a higher SNR compared with the SNR of the DW [diffusion-weighted] MR images obtained with conventional phase encoding; we speculate that this finding was, at least in part, attributable to the shorter echo time that was achieved with the reduction of phase-encoding steps at DW MR imaging with SENSE."
The 30% lower SNR that was calculated for the "conventional" SNR measurement in our study is easy to explain. In SENSE imaging, the background noise is eliminated but not on all parts of the image. The SENSE algorithm leaves a rim of about 10 mm of the actual background noise around the subject. This is where the ROI for background noise had to be placed. Since this is very close to the subject, physiologic noise will add to the noise level.
|
REFERENCE
|
|---|
- Kuhl CK, Gieseke J, von Falkenhausen M, et al. Sensitivity encoding for diffusion-weighted MR imaging at 3.0 T: intraindividual comparative study. Radiology 2005;234:517–526.[Abstract/Free Full Text]
TOP
References
REFERENCE 
