Parametric Imaging of [11C]Flumazenil Binding in the Rat Brain

Purpose

This study evaluates the performance of several parametric methods for assessing [¹¹C]flumazenil binding distribution in the rat brain.

Procedures

Dynamic (60 min) positron emission tomography data with metabolite-corrected plasma input function were retrospectively analyzed (male Wistar rats, n = 10). Distribution volume (V _T) images were generated from basis function method (BFM), Logan graphical analysis (Logan), and spectral analysis (SA). Using the pons as pseudo-reference tissue, binding potential (BP _ND and DVR–1) images were obtained from receptor parametric imaging algorithms (RPM and SRTM2) and reference Logan (RLogan). Standardized uptake value images (SUV and SUVR) were also computed for different intervals post-injection. Next, regional averages were extracted from the parametric images, using pre-defined volumes of interest, which were also applied to the regional time-activity curves from the dynamic data. Parametric data were compared to their regional counterparts and to two-tissue compartment model (2TCM)-based values (previously defined as the model of choice for rats). Parameter agreement was assessed by linear regression analysis and Bland-Altman plots.

Results

All parametric methods strongly correlated to their regional counterparts (R ² > 0.97) and to the 2TCM values (R ² ≥ 0.95). SA and RLogan underestimated V _T and BP _ND (slope of 0.93 and 0.86, respectively), while SUVR-1 overestimated BP _ND (slope higher than 1.07 for all intervals). While BFM and SRTM2 had the smallest bias to 2TCM values (0.05 for both), ratio Bland-Altman plots showed Logan and RLogan displayed relative errors which were comparable between different regions, in contrast with the other methods. Although SUV consistently underestimated V _T, the bias in this method was also constant across regions.

Conclusions

All parametric methods performed well for the analysis of [¹¹C]flumazenil distribution and binding in the rat brain. However, Logan and RLogan slightly outperformed the other methods in terms of precision, providing robust parameter estimation and constant bias. Yet, other methods can be of interest, because they can provide tissue perfusion (i.e., K ₁ with BFM and SA), relative flow (i.e., R ₁ with RPM and SRTM2), and model order (SA) images.