Characterization and Mitigation of Bias in Parametric Mapping of Reward-Induced Dopamine Release using Simultaneous Positron Emission Tomography and Magnetic Resonance Imaging

Abstract

The dopamine system plays an essential role in learning associated with reward expectation and receipt. [11C]Raclopride positron emission tomography (PET) can be used to observe dopamine D2/D3 receptor binding and its changes in response to the administration of a rewarded behavioral task. Binding potential can be measured by fitting a compartmental model to the PET time activity concentration in a region and task-related changes in binding potential (ΔBPND) can be measured by extending the model accordingly. However, measures of ΔBPND can be biased by undesired corrupting influences. In this work, two sources of bias in ΔBPND were characterized and mitigated: motion bias and model bias. Motion bias in ΔBPND was characterized by measuring the extent of the motion present in the data and performing simulations to estimate its effect both at different levels of task response and with different PET time framings. Increased motion was found to increase bias in ΔBPND. The greatest bias was observed at the edges of the striatum, where regions of high binding border regions of low binding. This bias becomes more confounding as the magnitude of the task response effect decreases. It can, however, largely be corrected by either frame-based motion correction or reconstruction-based motion correction. These two approaches perform comparably because the majority of the ΔBPND bias attributable to motion is caused by motion between frames rather than within frames. Model-based bias in ΔBPND arises when the compartmental model used to fit the PET data makes invalid assumptions (e.g. when data best described with two tissue compartments is fit with a model with one tissue compartment). Simulations were used to demonstrate that this bias is dependent on the underlying value of k4. A deweighting approach was applied to reduce model bias, decreasing extrastriatal bias and decoupling estimated ΔBPND from k2’ selection. Limitations of this method include increased overall noise and a modest bias reduction effect in the striatum. These sources of bias in ΔBPND should be accounted and corrected for when making associations between PET measures of binding potential and behavioral measures of rewarded learning.