Improving the measurement of the 99Tc(m)-ECD brain perfusion index by temporal analysis

Matsuda et al. have described a non-invasive method for brain perfusion quantification by computing the ratio of the cumulated counts in the cerebral hemispheres and aortic arch. The regions of interest (ROIs) are drawn manually and are observer dependent. The aim of this study was to develop a new method designed to minimize the intra- and interobserver variability when drawing the different ROIs. A dynamic study was performed as in Matsuda's method on 30 patients using technetium-99m ethyl cysteinate dimer (99Tc(m)-ECD) (ID: 800 MBq+/-33 MBq). The manual method of drawing ROIs was then compared with the following, automated one. A temporal analysis was performed on the cardiac first-pass study to obtain parametric images of the thorax. An ROI of the aortic arch was drawn automatically by means of an isocontour algorithm on the resulting views. The whole sequence was reframed and filtered by a temporal low-pass filter. Hemispheric brain ROIs were delineated on a summed image. Matsuda's algorithm was then applied. Intraobserver variability was evaluated for the classical Matsuda method. The correlation in brain perfusion index (BPI) measurements was r = 0.8976 for naive observers and r = 0.9443 for well-trained observers. Interobserver variability was also evaluated; the correlation was r = 0.7574 for naive observers and r = 0.9190 for well-trained observers. With our proposed method, the correlation in the measurements of BPI for evaluating the intraobserver variability was r = 0.9955 for naive observers and r = 0.9989 for well-trained observers. For interobserver variability, the results were r = 0.9234 for naive observers and r = 0.9230 for well-trained observers. We conclude that temporal analysis allows brain perfusion to be measured in a semi-automatic manner, and improves the reproducibility compared with the original method of Matsuda, particularly for naive observers.