Fig. 2.

In the line forward projection, voxels that contribute to LOR i are identified by performing an outer and an inner loop. The former iterates over the main dimension for the LOR (as defined in (7)—here e_x), while the latter iterates over the two remaining dimensions (only e_y is shown on the figure). The computation of the inner loops is done simultaneously in parallel shaders within the GPU. To make computation efficient, the inner loop bounds are increased (see Fig. 3) so that the number of iterations is constant. In a second pass, the outer loop sum is computed by a second shader (bottom).