Scatter correction of the projection images used to reconstruct volume CT images is a computationally intensive process that cannot at present be done in real time. We note, however, that although scatter is non-uniform, it varies smoothly - i.e. with a low spatial frequency. Existing Monte-Carlo estimation methods need to run over a large number of iterations in order to eliminate the mainly high-frequency artefacts that are present initially, but this process can be cut short and a low-pass filter applied instead thereby substantially reducing the computational load. By applying the same principle in the spatial domain and interpolating some scatter images from adjacent and surrounding scatter images, the computational load can be reduced still further. Applying these techniques in an iterative process on graphics-processor-quality computational hardware enables real-time scatter correction and reconstruction of CT volume images.
