Optimisation of quantification for 18F-DPA-714 in the healthy human

95 Objectives \nTSPO has been identified as a suitable marker for inflammation that may occur in neurodegenerative diseases and [18F]DPA-714 is one of several second generation TSPO tracers used in PET imaging to follow neuroinflammation. Quantification of [18F]DPA-714 in PET faces several challenges, including partial volume (PV) effect, noise in the PET data, and tracer binding to the vascular wall. This work was based on the Lavisse et al. 2015 study of [18F]DPA-714 PET binding parameters in healthy human subjects (Lavisse et al. JNM, 2015). Using this study as a basis we investigated improved reconstruction and image processing methods to determine whether the two tissue compartmental model (2TCM) or a 2TCM with an extra compartment included for vascular trapping of the ligand (2TCM-1K) was more appropriate for this data as described by Rizzo et al. (Rizzo et al. JCBFM, 2014). Methods \nEight healthy human subjects (5 high affinity binders, HAB, 3 mixed affinity binders, MAB) underwent [18F]DPA-714 PET in a 3D high resolution research tomograph (HRRT, Siemens, Knoxville, TN, USA) with arterial blood sampling. The scans were reconstructed with EMML and AB-EMML (a less biased method, as it allows for negative values) (Van Slambrouck et al. IEEE-MIC, 2015). The reconstructions had PSF modelling, were iterated to convergence and had a PSF smoothing applied (Stute et al. PMB, 2013). Both sets of reconstructions then had a 4D iterative deconvolution and spatiotemporal regularisation process (ID-SR) applied for PV correction and noise removal (Reilhac et al. Neuroimage, 2015). The two kinetic models mentioned (2TCM and 2TCM-1K) were applied, both using a metabolite corrected arterial input function. Results \nReconstruction: The volume of distribution (Vt) and binding potential (BP) values estimated using EMML and AB-EMML showed little difference and were strongly correlated (r2=0.997). The standard error (SE) for each parameter was typically lower by 2% using AB-EMML, so the images using this reconstruction were used for the following results. The 2TCM and 2TCM-1K models both fit well on the data, with the Akaike criterion (AIC) favoring the 2TCM-1K (62.5% of curves with lower AIC). The endothelial binding rate constant (kb) from the 2TCM-1K was extremely variable through the different regions and within the regions, with an average vallue of 0.61±1.08 for HABs and 0.65±1.01 for MABs.\nImpact of ID-SR: After applying the ID-SR, the 2TCM-1K model was shown as the most appropriate out of the two applied, with 100% of curves having a lower AIC than the 2TCM. The parameter estimates were much better identified, with SE values changing from 16% to 7% for Vt and from 49% to 21% for BP. The relative standard deviation for parameter estimates for each region also decreased for both the HAB group: Vt 60% to 55% and BP 77% to 54% and MAB: Vt reduced from 12% to 8% and BP 88% to 35%. The typical value for kb was reduced and more stable than before processing, at 0.22±0.11 (HAB) and 0.13±0.07 (MAB).\nModelling: The 2TCM-1K model was chosen as the most appropriate model based on visual assessment and AIC. The regional estimates for the AB-EMML reconstructions with ID-SR varied from 2.31 to 3.63 mL/cm3 for Vt, and 1.76 to 2.34 for BP in the HAB group, and from 1.80 to 2.99 mL/cm3 for Vt, and 1.38 to 1.60 for BP in the MAB group. The Vt and BP values were significantly lower (p Conclusions \nThe appropriate model was found to be the 2TCM-1K and using ID-SR improved the identifiability of the model and the variability of the parameters, strongly elucidating the endothelial binding compartment. A small improvement was found in parameter identifiabillity by using AB-EMML reconstruction.\n$$graphic_C7EBD081-51B9-4A31-B63E-88EC4CA8FC5E$$