A method to reduce bias in iterative scatter correction estimate for PET

Scatter corrections plays a critical role in producing quantitative PET images. In the Siemens mCT Biograph, used as a model for this work, an iterative Single Scatter Simulation is the method implemented to produce an unbiased emission image for scatter simulation. In particularly challenging phantom studies, the scatter may not converge after the specified number of iterations, but may oscillate, slowly converging after many iteration. In the standard method used for scatter simulation, the reconstructed image (lRD from iteration (i) is used as the input image (lS_i) for scatter simulation to generate the scatter sinogram. The scatter singoram is then used with the trues sinogram to generate the reconstructed image (IR_i+1) of the next iteration. The first two iteration of the proposed method operates the same way as the current method. At i>2, the reconstructed image (lR_i) is averaged with the previous input image (lS_i-I) to produce an average input image (lS_i) which will be used as an input image for the scatter simulation. In this study, we explain the proposed method and verify that the scatter fraction converges in a reasonable number of iterations. We will present investigation of scatter fraction and image quantification for phantom and patient data.