Usefulness of voxel-wise Pearson correlation coefficient for reproducibility and repeatability measurements in PET

Ziel/Aim:

In the field of molecular imaging using Positron Emission Tomography (PET), a method for measuring reproducibility or repeatability of serial PET is discussed continuously. For example the limited reproducibility of [¹⁸F]fluoromisonidazole PET measurements was shown using voxel-wise Pearson correlation coefficient (PCC) (1). But the applied method may not be suitable for this purpose (2). The argumentation follows the statement of Westgard and Hunt (3): “The correlation coefficient [...] is of no practical use in the statistical analysis of comparison data.” This investigation focuses on visualizing the usefulness of the PCC method.

Methodik/Methods:

A PET phantom was used to gather 3D volumetric PET data sets. The phantom consists of a cylinder filled with [¹⁸F]fluorodeoxyglucose. It contains four wax spheres partly made of [⁶⁸Ga]gallium-chloride and two glass spheres filled with a [⁶⁸Ga]gallium-chloride solution. After scanning this phantom in a clinical PET/CT scanner (Biograph16, Siemens, Germany) for 10 hours several 3D data sets were reconstructed showing the phantom at different contrast levels. PCC was then measured in pairs of high- and low-contrast data sets and in several sub volumes of the phantom: a) at the centre and b) at the boundary of all spheres, c) in the cylinder, d) on the cylinder-air boundary and e) in the air.

Ergebnisse/Results:

Even though all objects appear reproducible by visual interpretation, PCC inside the glass spheres, in the cylinder and in the air (0.13<0.91) was always lower than in wax spheres showing inhomogeneous activity distribution (0.92<1). PCC at object boundaries and in the wax spheres always indicated strong correlation (0.83 < PCC < 1). PCC in all target spheres was higher in high-contrast data sets compared to low-contrast data sets.

Schlussfolgerungen/Conclusions:

The fact that PCC is higher in volumes where activity gradients are located compared to homogeneously distributed volumes indicates that this method is not applicable to measure repeatability or reproducibility of PET measurements. PCC describes a combination of image properties like presence of signal gradients, signal homogeneity, noise and other imaging related effects.

Literatur/References:

1. Nehmeh SA, Lee NY, Schröder H, et. Al. (2008) Reproducibility of intratumor distribution of (18)F-fluoromisonidazole in head and neck cancer. Int J Radiat Oncol Biol Phys. 70(1):235-42.
2. Schwartz J, Humm JL, Gonen M, et. Al. (2011) Repeatability of SUV measurements in serial PET. Med Phys. 38(5):2629-38.
3. Westgard JO, Hunt MR. (1973) Use and interpretation of common statistical tests in method-comparison studies. Clin Chem. 19(1):49-57.