Window selection for improved dual photopeak window scatter correction

We have investigated the width and placement of the windows for the dual photopeak window (DPW) scatter subtraction method, in order to obtain a method that is stable on multi-head SPECT systems and capable of providing a good scatter estimate for extended objects. Stability and noise were examined with experiments using a SPECT system, while Monte Carlo simulations were used to predict the accuracy of scatter estimates for a variety of objects and to guide the development of regression relations for various window pairs. With a 15% asymmetric photopeak window and a 5% lower window abutted at 7 keV below the peak, a method was developed to relate the scatter-to-total (STR) ratio to the lower window-to-total ratio (LTR) at each pixel. Calibration was done using images of point sources at various depths in an attenuator. Using the entire image and a small, centered region of interest (ROI) at each depth, summed counts from the scatter component, the lower window, and the total window were used as input to a regression on a power function. The total image was multiplied by the STR to form the estimated scatter image to be subtracted. The DPW method with an extended total window demonstrated improved stability, achieved by abutting the two windows away from the peak. Noise in the scatter estimate was reduced by the use of a wider region of the spectrum, as well as by filtering. By including in the regression relation the additional information provided by the ROI, the bias of the estimated scatter from extended objects was reduced. For the simulated extended objects, DPW generally produced a less biased scatter estimate than the commonly used Compton window subtraction with k identical with 0.5.