1.2.1 - Measuring Image SNR

Ideally, a homogeneous material would have a uniform appearance on an MR image, i.e. all pixels corresponding to a uniform material would have the same grey-scale value. In practice, however, noise is always superimposed on the images, and is manifested in the image as a fluctuation of the pixel values around a central, average value for a uniform material. Noise in an image is most obviously apparent as a fluctuation of grey-scale pixel values in the image background where, under ideal circumstances, the signal intensity would be zero.

In a well-designed clinical MRI system, the noise in an image results mainly from the body itself. Because the human body is warm, it emits thermal energy at a wide spectrum of frequencies. This includes the well-known infrared energy that allows warm bodies to be imaged in the dark using infrared-sensitive detectors, as well as radiofrequency (RF) energy at the same frequency as the MR signal itself. This thermal radiofrequency energy

is "picked up" by the RF coil and fed back to the MRI scanner as thermal "noise". Thermal noise created by the body is thus superimposed on the actual MR signal, causing an uncertainty in the amplitude of the measured signal. Noise can also be introduced from the RF coil as well as other electronic components in the signal reception chain. These components must be carefully optimized so that noise arising from sources outside of the body itself is minimized. The image SNR is a ratio of the amplitude of the actual, desired MR signal from tissue to the noise in the image. Image quality is critically dependent on SNR; the SNR reflects how much certainty can be associated with the displayed signal values for the tissue that is being assessed. In general, it is desirable for the MR signal from tissue to be large compared to the size of the background signal fluctuations due to noise. The usual standard for measuring SNR in an image is the ratio of the mean signal intensity in the anatomic region of interest (ROI) to the standard deviation in a region in the background. Different tissues within the image will have different SNR values due to their different signal intensities, but the noise in the image is constant. Because the thermal energy produced by the body cannot be reduced, an MRI system must be optimized to provide the highest possible signal from tissue. The most critical design component of the imaging system for improving SNR in an image is the RF coil. The RF coil should be designed to "couple" strongly to the tissue being imaged, i.e. the magnetic interaction with the tissue of interest at the MR signal frequency should be maximized. Note that the magnetic coupling that occurs between the RF coil and patient will vary with the patient size and composition. The SNR for a particular tissue imaged using the same RF coil can therefore vary from patient to patient due to the different coupling characteristics of the coil with different patients and patient positioning relative to the coil. In general, a good RF coil is designed to provide high signal levels (i.e. sensitivity) from the tissue of interest over a wide range of patient sizes, and to create a minimal amount of additional extra noise from thermal energy of its own electronic components. The choice of pulse sequence and imaging parameters also affects the image SNR, and should be taken into consideration when designing protocols or assessing an image for diagnostic information. In the remainder of Section 1.2, the impact on SNR of imaging parameters common to all pulse sequences is discussed. In MR imaging, it is often necessary to strike a compromise between two counteracting image attributes, e.g. higher SNR can be achieved with any given protocol at the cost of increased acquisition time. It is important to understand how user-selectable parameters affect SNR, so that appropriate choices and trade-offs can be made.