ABSTRACT
Of course, this is rarely done except perhaps by a maniacal graduate student working on his or her dissertation and making measurement on a phantom rather that a patient.
Usually, to characterize the noise in an image, we obtain a single image that has a large region where the signal is uniform, and measure the mean and variance using multiple samples within this region. In radiography, this can be achieved by imaging a block of acrylic or some other uniformly thick material. In each case, the object placed in the x-ray beam mimicking a person is called a phantom. It is also a common practice to have nothing in the x-ray beam (except for air-hence the term “air-scan”) to obtain an image for such noise measurements. In each case, we assume that the signal is the same (except for random uctuations) at every location within the region of interest, and calculate the noise (standard deviation) using measurements at multiple sites within the region. is provides a good estimate of quantum noise assuming that there are no spatially correlated noise sources within the region of interest. A spatially correlated noise source might be generated by an oscillation in a video amplier, a dirty roller in the lm processor, or by some other (possibly unknown) process that aects neighboring values in the region of interest.
