Abstract
497
Objectives: Kinetic analyses of dynamic PET data with reversibly binding receptor-ligand radiotracers can be used to detect endogenous neurotransmitter (NT) releases elicited by cognitive tasks or drug stimulation. Conventionally, NT release is characterized by fitting time-activity curves (TACs) with an appropriate kinetic model such as LSRRM (Alpert et al 2003), an extension of the simplified reference region model that incorporates time-varying binding due to tracer displacement by NT. A limitation of this “indirect” approach, however, is the poor signal-to-noise ratio (SNR) of the estimated parametric images. In the current work, we developed a method for estimation of LSRRM parametric maps with significantly higher SNR by direct parameter estimation from raw dynamic PET projection data. We conducted simulation and human 11C-raclopride studies to evaluate the performance of this method against the standard indirect method. Methods: A numerical phantom comprised of 22 different brain regions including 7 striatal sub-divisions was created using the MNI atlas. Kinetic parameters values were assigned to each brain region based on measurements performed on 16 subjects studied with 11C-raclopride. Dopamine (DA) release was simulated in the striatum (executive area). LSRRM was used to generate the corresponding 4-D (i.e., 3-D + time) activity images which were then forward-projected to produce noise-free dynamic sinograms. Attenuation, detector sensitivity, point spread function and radioactive decay were modeled during sinogram data generation. After scaling to standard counts levels, Poisson noise was added to the sinograms to achieve noise levels comparable to our human studies. For empirical evaluation, a 45-min human study was conducted on the Siemens mMR camera using 11C-raclopride (13.5mCi). A reward task was started ~27 min after radiotracer injection to induce striatal DA release. For both simulation and human studies, parametric images of binding potential with non-displaceable reference (BPND) and magnitude of DA release (gamma) were computed using the indirect method and the proposed direct method. For the indirect approach, parameters were estimated by pixel-wise application of LSRRM using the cerebellum as a reference tissue input and weighted least-squares fitting of TACs obtained following fully-3D dynamic OSEM reconstruction. No post-reconstruction smoothing was included. For the direct method, parameters were directly estimated from the dynamic sinograms by preconditioned conjugate gradient based optimization of a four-dimensional Poisson log-likelihood objective function incorporating LSRRM kinetics and accounting for the effects of attenuation, sensitivity, scatter and randoms.
Results: In simulation studies, the indirect method estimated striatal BPND and gamma with 7.8% and 79.2% bias, respectively. For the direct reconstruction, bias was reduced to 6.9% and 5.5% for striatal BPND and gamma estimates, respectively. Coefficients of variation (‘CV’, i.e. pixel-wise variability) of striatal BPND and gamma estimates were 65.3% and 72.0% for the indirect method as compared to 17.4% and 17.9% for the direct method. Likewise, in the human study, the direct method yielded BPND and gamma images with improved SNR. CV of BPND and gamma estimates in the putamen were 71.4% and 91.4% for the indirect method and 32.9% and 37.6% for the direct method. Conclusion: Direct reconstruction can reduce bias and increase the SNR of parametric images. The method has the potential to improve the characterization of localized neurotransmitter release, potentially allowing detection of weaker effects, reducing the needed sample size, or reducing radioactivity dose to facilitate repeated experiments or minimize radiation exposure. Acknowledgment: This work was supported in part by NIH grants R01MH102279, R01MH100350 and P41EB022544.