Introduction
[11C](R)-PK11195 is a tracer of microglia activation in the brain. Most studies used a reference tissue model (RTM) for measuring activation 1 , but these models have not been validated against a plasma input model. The purpose of this study was to validate several RTMs against a previously described two tissue compartment plasma input model with fixed K1/k2 (PIM) 2 .
Methods
Five RTMs were evaluated: A. Two target and one reference tissue (full RTM) 3 ; B. One target and one reference tissue (simplified RTM) 4 ; C. Two target and two reference tissues; D. Model C with k4 = k4'; E. Three target and two reference tissues with k5 = k5' and k6 = k6'. Two methods were evaluated that obtained BPsp indirectly through Vd: F. Two target and one reference tissues; G. Two target and two reference tissues. One combined method was evaluated: H. BPsp from method B, corrected for BPns by multiplying with 1+BP'ns calculated using PIM. Simulations were performed to assess accuracy and precision of BPsp using clinically relevant parameter values, and varying BPsp, BPns and Vb. 500 runs per parameter set were performed. Noise was varied between 0 and 10% COV. Using clinical data, results from RTMs were compared with those from PIM.
Results
For standard parameter values (7.5% noise), simulations showed biases of −58%, −15% and −5% and variablities (COV) of 26%, 35% and 22% for methods B, D and H, respectively. Varying noise (0–10%) resulted in changing bias from −61 to −58% (method B), from 2.5 to 18% (method D) and from 0 to 8% (method H), and in reducing precision from 0 to 30%, 50% and 28% for methods B, D and H, respectively. Other methods provided poor accuracy and/or precision. Increasing BPns increased bias for methods B and D, but did almost not effect accuracy of method H. For small BPsp method D became unstable, whilst methods B and H still provided reasonable estimates. Clinical evaluation showed that methods A and B provided nearly identical results (R-squared = 0.94), but method B had better precision. Results from methods B, D and H correlated reasonably well with PIM (R-squared = 0.74, 0.65 and 0.84, respectively). Other methods resulted in very low correlations (R-squared <0.06), consistent with simulation results.
Discussion
Both simulations and clinical data showed that only methods B, D and H correlate with PIM. Apart from the best correlation, method H also showed the lowest bias at various simulated BPns, BPsp and Vb values. COV for both methods B and H were much lower than for method D. Method H allows for accurate and precise estimations of both BPsp and BPns. Method B is recommended when a plasma input is not available.