Smoothing of random data to reduce noise in 3-D PET

In the standard method of acquiring 3-D data on a CIT/Siemens 953B brain scanner operated with the septa in the retracted position, delayed events are used to estimate random events and are automatically subtracted from prompt events to give sinograms with true events. This technique of subtracting delayed events has the potential disadvantage of increasing the noise in the back-projected images due to the noise added to the final sinograms by the noise from the delayed events or random data. To reduce this source of noise, the authors have investigated the acquisition of the delayed and prompt data separately and then smoothing the random images prior to the subtraction from the prompt images. The random images were smoothed using a mean filter over a region of interest of 7×7 pixels. This technique has been tested as a function of random to trues ratio using a 20-cm uniform water phantom with F-18 activity. The images obtained through the use of smoothed random images subtracted from prompt images are shown to have less noise than true images obtained by subtracting unsmoothed random images from prompt images. However, even for the highest random to trues ratios commonly used (33%), this improvement is only about 6%