Optimal slice positioning for long bone CT reconstruction

The aim of this work is to optimally reconstruct the 3D geometry and density distribution of long bones starting from a limited number of CT scan slices, Nowadays CT slices positioning is assessed by the radiologist starting from a 2D prescanning image of the bone segment. In the proposed technique, the optimal scanning sequence is searched by minimizing the overall reconstruction error of the bone density distribution, predicted by linear interpolation between the slices selected, with respect to the true 2D distribution. The technique, based on an iterative discard-insert-exchange optimization method, has been tested on both a composite and a human femur CT dataset consisting of about 450 slices. For the composite femur, the 3D root mean square reconstruction error obtained with 50 slices by a radiological protocol was equal to 12.6 grey levels, on a 0-255 range, decreasing to 6.1 for the 2D-optimized protocol. Analogously, for the human femur 11.9 and 9.6 grey levels were found respectively. The procedure was validated by comparison with 3D optimization which gave only slightly lower errors, 5.7 and 9.1 grey levels respectively. It is concluded that 2D-optimization provides good sub-optimal scanning protocol for 3D bone reconstruction