Improved drug efficacy assessment using joint Bayesian estimation framework

Abstract The development of central nervous system (CNS) drugs requires imaging tools that can quantify drug efficacy and target engagement as well as the effects of drug concentration on these quantitative assessment measures. Positron emission tomography (PET) imaging has become standard practice for quantitative assessment, thanks to its ability to image tracers that bind to target receptors of drugs. Receptor occupancy (RO) studies, which consist in a pair of PET scans, baseline and post-drug injection, can be used to quantify, as a function of drug concentration, the blocking effect of a drug. The concentration yielding half-maximum blocking effect, called EC50, is determined by first estimating receptor occupancy for each pair of scans, then fitting a logistic model to the occupancy vs. concentration curve. The conventional method to estimate RO and EC50 consists in reconstructing the dynamic PET data for each pair of scans, perform kinetic fitting, typically in high binding regions, to estimate the binding potential and calculate RO, before performing the logistic fit across concentrations to obtain EC50. The resulting estimates have low precision due to the noise in dynamic PET images and the lack of proper noise modeling. Moreover, estimating single RO and EC50 values discards potentially valuable information on the spatial distribution of RO and EC50. We propose an estimation framework that jointly estimates spatial RO maps for each pair of scans and a global EC50 map using an end-to-end model from the PET measurements to the estimated EC50. The estimation framework relies on advanced optimization strategies to decompose the joint estimation process into manageable subproblems, such as image reconstruction, parametric fitting, image denoising and logistic fitting. The method is expected to improve the estimation of RO and EC50 over conventional methods, while offering additional spatial information. The targeted improvement in estimation would allow to reduce the sample size required in drug trials to achieve the same statistical power as conventional approaches (or conversely, increase the statistical power for a fixed sample size). The method will be validated in numerical simulations and applied to in vivo animal studies.