A reference tissue implementation of Simultaneous Multifactor Bayesian Analysis (SiMBA) of PET time activity curve...

PET analysis is conventionally performed as a two-stage process of quantification followed by analysis. We recently introduced SiMBA (Simultaneous Multifactor Bayesian Analysis), a hierarchical model that performs quantification and analysis for all brain regions of all individuals at once, and in so doing improves both the accuracy of parameter estimation and inferential efficiency. However until now, SiMBA has only been implemented for the two-tissue compartment model. We have now extended this general approach to also allow a non-invasive reference tissue implementation that includes both the full reference tissue model (FRTM) and the simplified reference tissue model (SRTM). Although the FRTM did not successfully converge and gave rise to unlikely parameter estimates for this application, the SRTM described the data well. In simulated data, SiMBA improves quantitative parameter estimation accuracy, reducing error by, on average, 57% for binding potential ( $B{P}_{\text{ND}}$ ) using SRTM. In considerations of statistical power, our simulation studies indicate that the efficiency of SiMBA modeling approximately corresponds to improvements that would require doubling the sample size if using conventional methods, with no increase in the false-positive rate. We applied the model to PET data measured with [¹¹C]AZ10419369, which binds selectively to the serotonin 1B receptor, in datasets collected at three different PET centres (n = 139, n = 44 and n = 39). We show that SiMBA yields replicable inferences by comparing associations between PET parameters and age in the different datasets. Moreover, we show that time activity curve data from different centres can be combined in a single SiMBA model using covariates to control between-centre parameter differences, in order to harmonise data between centres. In summary, we present a novel approach for non-invasive quantification and analysis of PET time activity curve data which improves quantification and inferences, enables effective between-centre data harmonisation, and also yields replicable outcomes. This method has the potential to significantly expand the range of research questions which can be meaningfully tested using conventional sample sizes with PET imaging.