Variation of Kinetic Model Parameters Due to Input Peak Distortions and Noise in Simulated 82Rb PET Perfusion Studies

Cardiac ⁸²Rubidium dynamic PET studies allow quantifying myocardial perfusion by using tracer kinetic modelling. The accuracy of the derived perfusion parameter K₁ is limited e.g. by noise, which impacts the measurement of both the myocardium time activity curve (TAC) and the input function. Measuring the input function peak is particularly challenging for the sharp injection bolus. By simulation studies with a two-compartment model, we quantified the impact of both input peak distortions and data noise on K₁. The input function was a generic analytical function. Modifying a function parameter, peak distortions were introduced. Gaussian noise was added to the simulated myocardium TAC and the model was fitted using the distorted input function. We analysed various sets of functions and model parameters covering a physiologically relevant range. We show that up to 20% noise in the tissue data, even moderate errors in the input peak (e.g. 10%) can dominate the error of the K₁ estimate. Generally, the bias due to the peak distortion and the variance due to the tissue noise can easily lead to more than 30% error in K₁. These data highlight the importance of optimizing the clinical acquisition protocols and data processing, especially with regard to a correct input peak estimation. An analysis similar to the one suggested here can help in this optimization, and can furthermore provide confidence limits on the perfusion parameters presented to the clinician.