Performance evaluation of various parametric methods: For [11C]PIB

Introduction: [11C]PIB is a ligand for imaging amyloid fibrils in brain. Price et al.[1] and Lopresti et al.[2] have evaluated Methods: for quantification of [11C]PIB studies, including performance of various parametric Methods:. The aim of the present study was to further investigate performance of parametric Methods: in which the reference tissue efflux rate (k2') is fixed to improve quantification. Performance of parametric Methods: was evaluated using simulations and clinical data. Methods: The following parametric Methods: were studied: RPM1[3], Reference Logan[4], MRTMo and MRTM[5]. In addition, RPM2[6], MRTM2[5] and two newly developed multi-linear Methods: (MRTM3 and MRTM4), with fixed k2', were included. Both simulations and clinical data were used to determine effects of flow (R1=0.6-0.9), fractional blood volume (Vb = 0.025-0.075) and binding potential (BP=0.2-3) on precision and accuracy of parametric BP and DVR. For simulations various [11C]PIB time activity curves (TACs) were generated at a 15% noise level using both reference tissue input (RI) and plasma input (PI). In addition, test-retest studies (two scans with arterial blood sampling on one day) were performed in 6 control and 5 AD subjects. TACs were derived from 8 regions of interest in the frontal cortex. Grey matter cerebellum was used as reference tissue. BP obtained with SRTM and DVR obtained with two-tissue reversible plasma-input model (2T-4k DVR) were used as reference. Results: RI simulations indicated that differences in bias between the various Methods: reduced with increasing BP (>0.2). Overall, lowest BP biases were observed for MRTM2 (<4.3±26%) and RPM2 (<7.4±24%), i.e. when k2' was fixed. Simulations at different flow settings (R1=0.6-0.9) indicated lowest BP biases for MRTM2 (<3±12%), RPM1 (<1±23%) and RPM2 (<4±11%). PI simulations indicated a similar performance compared with RI simulations. Increasing Vb increases bias for all Methods: For clinical data all Methods: showed good correlation with SRTM (R2=0.89-95). The correlation coefficient (R2) for RPM1 was 0.95 and it was 0.92 for MRTM, MRTM2 and MRTM3. Both simulations and clinical data indicated a larger bias (about -15%) for reference Logan, MRTMo, MRTM3, MRTM4 than for MRTM, RPM1, RPM2 and MRTM2 (bias <5%) compared with both BP-SRTM and/or 2T-4k DVR. The difference in average BP between AD and control subjects was (P<0.001), however, slightly smaller for RPM1, MRTM2 and RPM2 than for the other Methods:. Finally, best DVR test-retest variability was found for MRTM3 (2.7%), but was slightly worse for other Methods: (max 4.1%). Conclusions: MRTM, RPM1, RPM2 and MRTM2 were the most accurate parametric Methods: for [11C]PIB studies. Fixing k2′ had a small beneficial effect on the performance of these Methods:. Visual inspection of parametric images also revealed a small improvement of image quality using k2′ fixed, consistent with simulations. RPM2 is the recommended method for [11C]PIB.