Characterization and Mitigation of Model Bias in Parametric Mapping of Dopamine Response to Behavioral Challenge
Compartmental modeling of 11C-raclopride (RAC) is commonly used to measure dopamine response to intra-scan behavioral tasks. Bias in estimates of binding potential (BPND) and its dynamic changes (ΔBPND) can arise when the selected compartmental model deviates from the underlying biology. In this work, we characterize the bias associated with assuming a single target compartment and propose a model for reducing this bias by selectively discounting the contribution of the initial uptake period. Methods: 69 healthy young adult participants were scanned using RAC PET/MR while simultaneously performing a rewarded behavioral task. BPND and ΔBPND were estimated using an extension of the Multilinear Reference Tissue Model (MRTM2) with the task challenge encoded as a Heaviside step function. Bias was estimated using simulations designed to match the acquired data and was reduced by introducing a new model (DE-MRTM2) that reduces the biasing influence of the initial uptake period in the modeled estimation of BPND for both simulations and participant data. Results: Bias in ΔBPND was observed to vary both spatially with BPND and with the assumed value of k4. At the most likely value of k4 (0.13 min-1), the average bias and the maximum voxel bias magnitude in the nucleus accumbens were estimated to be 1.2% and 3.9% respectively. Simulations estimated that debiasing the contribution of the first 27 minutes of acquired data reduced average bias and maximum voxel bias in the nucleus accumbens ΔBPND to -0.3% and 2.4% respectively. In the acquired participant data, DE-MRTM2 produced modest changes in the experimental estimates of striatal ΔBPND, while extrastriatal bias patterns were greatly reduced. DE-MRTM2 also considerably reduced the dependence of ΔBPND upon the first-pass selection of k2'. Conclusion: Selectively discounting the contribution of the initial uptake period can help mitigate BPND- and k4-dependent bias in single compartment models of ΔBPND, while also reducing the dependence of ΔBPND on the first-pass estimation of k2'.