PET/CT Assessment of Response to Therapy: Tumor Change Measurement, Truth Data, and Error

Abstract

We describe methods and issues that are relevant to the measurement of change in tumor uptake of (18)F-fluorodeoxyglucose (FDG) or other radiotracers, as measured from positron emission tomography/computed tomography (PET/CT) images, and how this would relate to the establishment of PET/CT tumor imaging as a biomarker of patient response to therapy. The primary focus is on the uptake of FDG by lung tumors, but the approach can be applied to diseases other than lung cancer and to tracers other than FDG. The first issue addressed is the sources of bias and variance in the measurement of tumor uptake of FDG, and where there are still gaps in our knowledge. These are discussed in the context of measurement variation and how these would relate to the early detection of response to therapy. Some of the research efforts currently underway to identify the magnitude of some of these sources of error are described. In addition, we describe resources for these investigations that are being made available through the Reference Image Database for the Evaluation of Response project. Measures derived from PET image data that might be predictive of patient response as well as the additional issues that each of these metrics may encounter are described briefly. The relationship between individual patient response to therapy and utility for multicenter trials is discussed. We conclude with a discussion of moving from assessing measurement variation to the steps necessary to establish the efficacy of PET/CT imaging as a biomarker for response.

Figure 1

Sample images from the RIDER PET/CT collection. Serial coronal sections of PET/CT images of a patient with lung cancer (arrows).

Figure 2

(A) Global calibration factor for a clinical PET/CT scanner during an 18-month period. Indicated is an erroneous calibration factor that was used for approximately one month before detection. Values are indicated for both the standard ¹⁸F-FDG calibration method and using long-lived ⁶⁸Ge sources showing significant variability. (B) FDG PET images using the median (left image) and the outlier (right image) global calibration factors. The images are identical except for an overall scale difference in all SUVs of approximately 25%. This type of periodically vulnerable calibration error has clear consequences for longitudinal studies.

Figure 3

Partial volume and variance effects for a single scanner. (A) CT and PET images of the SNM validation phantom. (B) Absolute recovery coefficient as a function of sphere diameter and method of reporting SUV from an ROI placed over the spheres.

Figure 4

Scanner calibration factors (relative to mean value) during a 4-year period. Datawere collected with manufacturer-recommended procedures using ¹⁸F in a water-filled cylinder (diamonds) and the same ⁶⁸Ge in epoxy cylindrical source.

Figure 5

Sample calculation of change metrics from serial PET/CT image sets illustrating a potential use of the RIDER collection. Each symbol represents the change values for an individual lesion for the difference metrics.

Figure 6

Illustration of sample size versus minimum expected effect size for different total noise in measurements for a test power of 80% and significance level (α) = 0.05 (adapted from Doot et al. [47]). Total noise includes, for example, biologic, local calibration, and multicenter measurement effects.