PT  - JOURNAL ARTICLE
AU  - Alvin Ihsani
AU  - Arkadiusz Sitek
AU  - Yoann Petibon
AU  - Chao Ma
AU  - Paul Han
AU  - Georges El Fakhri
AU  - Jinsong Ouyang
TI  - Markov Chain Monte Carlo Estimation of Non-stationary PET Kinetic Parameters Compartment Models: A Flow Phantom Study
DP  - 2018 May 01
TA  - Journal of Nuclear Medicine
PG  - 1721--1721
VI  - 59
IP  - supplement 1
4099  - http://jnm.snmjournals.org/content/59/supplement_1/1721.short
4100  - http://jnm.snmjournals.org/content/59/supplement_1/1721.full
SO  - J Nucl Med2018 May 01; 59
AB  - 1721Introduction: The quantitation of the impact of a stimulus is of great interest in both clinical and pre-clinical Positron Emission Tomography (PET) imaging. Such studies vary from drug-challenge experiments (e.g. quantifying the impact of a drug in receptor binding in the brain) to rest-stress flow studies (e.g. quantifying degree of ischemia in myocardial tissue). Stimulus quantitation studies are typically performed in two phases: (1) the physiological state is estimated before the application of the stimulus, (2) the physiological state is estimated after the physiological state reaches a new steady state after the stimulus is applied. The conventional approach to estimating kinetic parameters in non-stationary studies involves performing non-linear least-squares (NNLS) curve-fitting of the kinetic model to each part of the non-stationary study. As an example, consider a rest dynamic PET myocardial perfusion imaging (MPI) where pharmacologic stress is applied to identify reversible perfusion defect during the “stress” phase. For such studies, the TAC after the stimulus is no longer independent from pre-stimulus due to the background activity left over from pre-stimulus as well as the temporary non-steady state during the stimulus. In this case, the NNLS approach can lead to inconsistent results across repeated experiments (e.g. increased bias). We propose a Markov Chain Monte Carlo (MCMC) approach to estimate kinetic parameters in non-stationary studies that provides superior fits, smaller bias, estimates the distribution of the kinetic parameters, and can capture the behavior of the kinetic parameters during the transition between steady states. Experiment Setting A dynamic kidney dialysis phantom was connected to a pump, which was used to change the flow. Four dynamic PET experiments were performed as listed in Table 1. In experiments 1 and 2 the flow was kept constant and a single dynamic PET (i.e., a bolus injection and list-mode acquisition) was performed. In experiments 3 and 4, the PET acquisition was first performed at a constant flow, the flow was changed gradually during the acquisition, and the acquisition continued for after the flow became stable. A bolus of [18F]FDG was administered at the beginning of all experiments, and again after the flow became stable in experiments 3 and 4. Due to the agglutinative nature of FDG, we found that a two-tissue non-stationary compartment model (see Figure 1) described the kinetics of FDG in the phantom more adequately. View this table:Table 1: A list of the acquisitions performed on the kidney dialysis phantom. Results: Figure 2 shows the distribution of the kinetic parameter k1 (a parameter of interest in flow studies) in experiments 1 and 2 along with the best fits of the MCMC and NNLS estimators. An important observation is that the NNLS k1 are within the 95% credible intervals of the MCMC-estimated k1 distributions which makes the results statistically indistinguishable. Figure 3 compares k1 as derived by each method in all experiments. It is apparent that while there is bias in the k1 derived from non-stationary studies, the MCMC estimator has less bias than NNLS. The proposed MCMC approach can also be used to identify parameters that do not change in a non-stationary study. We assumed that k3 and k4 are also non-stationary, however, the distribution of these parameters does not change significantly before and after the “stimulus” (see Figure 4), whereas k1 and k2 clearly show displacement before and after the “stimulus”, and are therefore truly non-stationary. Summary There is great potential in employing an MCMC estimator in non-stationary studies, because we obtain much more information about the parameter space. The MCMC estimator provides accurate bounds on the kinetic parameters, and may be used as a tool in model refinement as it can show which parameters are indeed non-stationary. Most importantly, the MCMC estimator is able to capture the transient state (e.g. between 3-4 min) where the flow is gradually changing during the experiment. Support: NIH P41EB022544