TY - JOUR T1 - The impact of parameter bound selection in the simplified reference tissue method on the accuracy and quality of parametric images JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 359 LP - 359 VL - 58 IS - supplement 1 AU - Ivan Klyuzhin AU - Nasim Vafai AU - Elham Shahinfard AU - Vesna Sossi Y1 - 2017/05/01 UR - http://jnm.snmjournals.org/content/58/supplement_1/359.abstract N2 - 359Objectives: In PET tracer kinetic modeling (KM), the simplified reference tissue method (SRTM) proposed by (Gunn et al., 1997) and SRTM2 developed by (Wu and Carson, 2002) are widely used to compute parametric images of the non-displaceable binding potential (BPND), tissue efflux rate k2 and tissue to reference influx ratio R1. The fitting parameter θ3 of the model is optimized by exhaustive direct search within pre-defined parameter bounds (PBs). The originally suggested bounds were [10-3, 10-2] s-1. The objective of this work was to investigate how the choice of PBs on θ3 may affect the accuracy and quality of SRTM2 parametric images for two tracers: [11C]dihydrotetrabenazine (DTBZ), a marker of VMAT2, and [11C]raclopride, a D2 receptor antagonist.Methods: Ten healthy control (HC) and 37 Parkinson’s disease (PD) subjects (reduced DTBZ binding) underwent 60-minute scans with DTBZ and RAC on a high-resolution research tomograph (HRRT, Siemens). The coincidence data were binned into a dynamic frame sequence (4×1 min, 3×2 min, 8×5 min, 1×10 min), reconstructed using OSEM-OP (voxel size [1.2 mm]3), and smoothed using a 2.0-mm Gaussian filter. Parametric images were produced using SRTM2 as implemented in the Receptor Parametric Mapping software. The occipital cortex was used as the reference with DTBZ, and cerebellum was used with RAC. Three PBs on θ3 were tested: [10-3, 10-2] (PB1), [10-4, 10-2] (PB2) and [0.5x10-2, 2.5x10-3] s-1 (PB3). The model fit quality was assessed by comparing the measured voxel time-activity curves (TACs) to their respective fits. The mean BPND in striatal regions of interest (ROI) was compared between the PBs. The ROIs were obtained by segmenting the putamen and caudate in MRI images that were acquired for each subject and co-registered to the PET images.Results: Using PB1 with DTBZ resulted in the lower bound θ3=10-3 producing the best fit for a large fraction of voxels in high uptake regions. Examination of measured and fitted voxel TACs revealed that the fits under-estimated activity values in later frames. With RAC, the fits followed the measured TACs adequately. Using PB2 produced more accurate DTBZ fits in general, however for some voxels the fits were strongly skewed by noise/outliers in the TACs. This introduced a characteristic high-frequency noise pattern in the DTBZ and RAC parametric images, with BPND (k2) ranging from -0.7 to ~13 (~0.001 to ~0.01). Through manual refinement, PB3 was found to offer a reasonable trade-off between the fit fidelity and noise. It was determined that with PB2, the noise could be eliminated by de-noising the voxel TACs (ref) prior to SRTM2. The mean DTBZ BPND in the posterior putamen (caudate) of the HC subjects was 2.71±0.40 (2.62±0.28) with PB1, and 3.12±0.50 (2.86±0.24) with PB3 (for 3 HC subjects the increase in BPND was 0.6-0.7). In early PD subjects (duration <= 2 y, N=9), the caudate BPND was 1.92±0.33 with PB1 and 2.08±0.41 with PB3. The noise pattern observed with PB2 affected the mean DTBZ and RAC BPND values by 5-20%.Conclusion: The originally proposed PBs in SRTM/SRTM2 may under-estimate BPND for tracers with relatively high binding potentials (e.g. DTBZ). On the other hand, widening the PBs by an order of magnitude makes the fits prone to noise/outliers in the voxel TACs, thereby introducing high-frequency noise in the parametric images. This highlights the need to adjust the PBs according to the tracer, administered dose and scanner resolution, with possible application of de-noising. ER -