Enhanced parameter estimation from noisy PET data: Part II--evaluation.

Abstract

RATIONALE AND OBJECTIVES: Positron emission tomography (PET) is a minimally invasive imaging modality that provides three-dimensional distribution data for a radioactive tracer concentration within the body. Local functional parameters are estimated from these images by fitting tracer kinetic data with mathematical models. However, in some applications, the reliability of parameter estimates may be hindered by the presence of noise. In the accompanying report in this issue of Academic Radiology, a novel method using principal component analysis (PCA) was presented and used for deriving parametric images of lung function from imaged tracer kinetics of intravenously injected nitrogen 13 (13NN) in saline solution. The PCA method averages 13NN concentrations from groups of voxels (volume elements) selected for their similarity in kinetics, rather than their spatial proximity. The goal of this study is to conduct a Monte Carlo simulation to evaluate the robustness to noise of parameters derived by means of the PCA method. MATERIALS AND METHODS: This evaluation involved: (1) generating "noise-free" synthetic PET images from experimental PET data, (2) adding noise to these images, (3) applying the PCA method to yield parametric images, and (4) comparing these parametric images with original noise-free images. RESULTS: Local parameters recovered by using the PCA method deviated from noise-free parameters on average by less than 1% for up to 32-fold of expected noise levels. These deviations were much less than those (>10%) recovered by using a direct curve-fitting method. CONCLUSION: The novel PCA approach provides robust parametric lung functional images while preserving the spatial resolution of the original images.