An artificial neural net and error backpropagation to reconstruct single photon emission computerized tomography data

A new radiographic patient positioning technique developed for radiosurgery has been analyzed to show the effect of computed tomography (CT) slice thickness on the precision of target localization during treatment. The positioning technique establishes the pose of the patientʹs anatomy during treatment by comparing treatment room radiographs with digitally reconstructed radiographs derived from a CT study. The measured pose is then used to align the x-ray therapy beam with the treatment site, without resorting to mechanical fixation. The technique has been found to be sensitive to submillimeter changes in skull position, which is the level of precision desired for radiosurgery. In this report it is shown that the precision of head localization improves by a factor of 2 when the CT slice t0hickness is reduced from 3.0 to 1.5 mm. This indicates that, in radiosurgical applications, image-guided beam alignment can be significantly influenced by the spatial resolution of the reference CT study. This result is relevant to all high-precision radiographic positioning techniques that utilize CT images. © i999 American Association of Physicistx in Medicine. [S0094-2405 (99)00102-9]