Measurement of Signal-to-Noise Ratio and Parallel Imaging

Abstract

The signal-to-noise ratio (SNR) of an image is a fun-damental quantitative measure of MR image quality and system performance, and has enormous impact on the diagnostic quality of clinical studies. The SNR measurements provide a direct means for com-parison of signal levels between different imaging method, patients, reconstruction methods, coils, and even different scanner systems. It is one of the essen-tial measures of system performance and is used rou-tinely in quality assurance protocols that monitor scanner performance (Price et al. 1990). The advent of parallel imaging with the develop-ment of SMASH (Sodickson and Manning 1997) and SENSE (Pruessmann et al. 1999), as well as other parallel imaging methods (Griswold et al. 2000, 2002; McKenzie et al. 2002; Kellman et al. 2001; cf. Chap. 2), has had tremendous impact on modern clini-cal scanning protocols. Parallel acceleration methods are commonly used to reduce acquisition time, with the trade-off that SNR of the resulting images will be reduced. For this reason, the ability to make accurate SNR measurements to measure the performance of parallel imaging applications is even more essential. Unfortunately, routine methods commonly used to measure SNR with single coil imaging applications are no longer valid when using parallel imaging methods. Noise becomes distributed unevenly across the image (as discussed in Chap. 3) and application of routine measurement methods can lead to highly erroneous SNR measurements. Great caution must be used when measuring SNR with parallel-imaging applications. In this chapter we review correct methods for the measurement of SNR in conventional single coil images, and review the underlying assumptions that are made. The concept of local noise amplifi cation, characterized by the geometry, or " g-factor " is then discussed, and reasons why the application of conventional methods for SNR measurement will fail when applied to images acquired with parallel imaging. Finally, three methods for SNR measurement that are compatible with paral-lel-imaging applications are discussed.