A study of shape-dependent partial volume correction in pet imaging using ellipsoidal phantoms fabricated via rapid prototyping

Positron emission tomography (PET) with 2-[18F]fluoro-2-deoxy-D-glucose (FDG) is being increasingly recognized as an important tool for quantitative assessment of tumor response because of its ability to capture functional information about the tumor's metabolism. However, despite many advances in PET technology, measurements of tumor radiopharmaceutical uptake in PET are still challenged by issues of accuracy and consistency, thereby compromising the use of PET as a surrogate endpoint in clinical trials. One limiting component of the overall uncertainty in PET is the relatively poor spatial resolution of the images which directly affects the accuracy of the tumor radioactivity measurements. These spatial resolution effects, colloquially known as the partial volume effect (PVE), are a function of the characteristics of the scanner as well as the tumor being imaged. Previous efforts have shown that the PVE depends strongly on the tumor volume and the background-to-tumor activity concentration ratio. The PVE is also suspected to be a function of tumor shape, although to date no systematic study of this effect has been performed. This dissertation seeks to help fill the gap in the current knowledge about the shape-dependence of the PVE by attempting to quantify, through both theoretical calculation and experimental measurement, the magnitude of the shape effect for ellipsoidal tumors. An experimental investigation of the tumor shape effect necessarily requires tumor phantoms of multiple shapes. Hence, a prerequisite for this research was the design and fabrication of hollow tumor phantoms which could be filled uniformly with radioactivity and imaged on a PET scanner. The phantom fabrication was achieved with the aid of stereolithography and included prolate ellipsoids of various axis ratios. The primary experimental method involved filling the tumor phantoms with solutions of 18F whose activity concentrations were known and traceable to primary radioactivity standards held by the National Institute of Standards and Technology (NIST). The tumor phantoms were then placed inside a Jaszczak cylinder (representing the human body) and imaged on a PET scanner located at NIST. This experimental approach allowed for the testing of: (1) The relative difference between tumors phantoms of different shapes, but same volume; (2) The overall accuracy of the PET measurements in terms of a ground truth reference value. Theoretical calculations of the tumor shape effect were also performed by mathematically convolving the phantom shapes with a 3D Gaussian point-spread function, and the results of the calculations were compared with the experimental data. The data show that the shape effect in PET tumor imaging can be as large as 15% for ellipsoid phantoms with axis ratios of 2:1, volume of 1.15 cm 3, and tumor-to-background activity concentration ratio of 9:1. This is explained by a greater loss of counts along the minor axis direction in the ellipsoid tumors compared to that of spheres of the same volume. The results of this PhD research confirm the existence of a tumor shape effect PET imaging. However, except in the case of ellipsoids with major-to-minor axis ratio greater than 2:1, a correction for the effect using recovery coefficients is expected to be challenging because its magnitude is comparable to the repeatability of the PET measurements.