Error analysis of the double-integral method for calculating brain blood perfusion from inert gas clearance data

A single-photon dynamic computer-assisted tomograph (DSPECT) has been built and is currently being used to evaluate regional cerebral blood perfusion in patients and volunteers. A computer simulation of the system was created to analyze the effects of data collection, Poisson noise, attenuation compensation, and the reconstruction technique now employed in the DSPECT. Several methods of attenuation compensation were used to generate perfusion images from both ideal and noisy data. The results indicate that the mean perfusion is calculated to within 10.4% accuracy for all perfusion rates in a region of interest if attenuation correction is used. Without attenuation correction, perfusions are underestimated by as much as 27%. The three correctors tested have different effects on the calculated perfusion value, depending on the location of the region of interest in the picture. The algorithm introduces random noise that is proportional to both the random error in the input data and the perfusion rate. Air-curve delay errors result in inaccuracies in the final perfusion picture that are proportional to perfusion rate. Physiological values (0.8-1.5) of the partition coefficient cause overestimation of both gray (0-34%) and white (7-67%) matter perfusion values. Compton scatter and collimator effects were not addressed in this study.