The Precision of Positron Emission Tomography: Theory and Measurement

Publisher Website
Google Scholar
Abstract
Summary: The limits of quantitation with positron emission tomography (PET) are examined with respect to the noise propagation resulting from radioactive decay and other sources of random error. Theoretical methods for evaluating the statistical error have been devised but seldom applied to experimental data obtained on human subjects. This paper extends the analysis in several ways: (1) A Monte Carlo method is described for tracking the propagation of statistical error through the analysis of in vivo measurements; (2) Experimental data, obtained in phantoms, validating the Monte Carlo method and other methods are presented; (3) A difference in activation paradigm, performed on regional CBF (rCBF) data from five human subjects, was analyzed on 1.6-cm diameter re- At an elementary level of description, positron emission tomography (PET) estimates the radioactivity concentration in one or more slices through a three-dimensional object. Many factors influence the accuracy and precision of the local concentration measurements with PET. Systematic errors (deterministic inaccuracies) arise from the finite resolution of the PET scanner, both in-plane and axial, and from the coarse sampling of slice-by-slice measurements. The limitations imposed by incomplete three-dimensional sampling are not widely appreciated. Another class of systematic errors arises from scanner calibration and corrections due to factors such as photon attenuation, scattered coincidences, and scanner deadtime. The precision of the gions of interest to determine the mean fractional statistical error in PET tissue concentration and in rCBF before and after stereotactic transformation; and (4) A linear statistical model and calculations of the various statistical errors were used to estimate the magnitude of the subject-specific fluctuations under various conditions. In this specific example, the root mean squared (RMS) noise in flow measurements was about three times higher than the RMS noise in the concentration measurements. In addition, the total random error was almost equally partitioned between statistical error and random fluctuations due to all other sources.