Page 203 - 89Zr-Immuno-PET:Towards a Clinical Tool to Guide Antibody-based Therapy in Cancer

Measurement variability due to noise is significant in 89Zr-immuno-PET
Chapter 5 evaluates noise-induced variability of 89Zr-immuno-PET for quantification of uptake in normal tissues and tumors. Per original scan, raw PET data was split in two equal parts and reconstructed into two count-reduced images (each representing 50% of the original injected dose). Noise-induced variability (expressed as the within-subject coefficient of variation (CoV)) for 89Zr-antiCD20, 89Zr-antiEGFR and 89Zr-antiCD44 was ~3% for large organs (brain, liver, lung, kidney, spleen) and ~20% for tumors. For tumor quantification, this is a significant source of measurement variability.
Interobserver reproducibility of tumor quantification for 89Zr-immuno-PET is excellent
Before tumor uptake quantification with 89Zr-immuno-PET can be evaluated in large clinical trials, multicenter reproducibility has to be assessed. In Chapter 6, interobserver reproducibility of tumor uptake quantification for 89Zr-immuno- PET was determined, using the same software and standard operating procedure in two centers (VUMC and UMCG). Three observers manually delineated tumor volumes of interest (VOI) (n=103) for 89Zr-antiCD20, 89Zr-antiEGFR and 89Zr- antiHER2. Maximum, peak and mean standardized uptake values (SUVmax, SUVpeak and SUVmean) were used to quantify tumor uptake.
Interobserver reproducibility (median CoV) was excellent for SUVmax and SUVpeak (0% and 0%). For SUVmean, measurement error due to interobserver variability in tumor uptake quantification was 7%. Even for SUVmax, there were cases where the three observers did not obtain exact the same value (26/103). This was due to insufficient contrast for manual delineation, resulting in delineation of a different structure (4/103) or location of the voxel with the maximum intensity at the border of the VOI (16/103). These tumor lesions cannot reliably be quantified and should therefore be excluded. Standardization by application of eligibility criteria for VOI quantification (e.g. VOI are deemed ineligible when the voxel with the maximum uptake was located at the edge of the VOI) is recommended to optimize interobserver reproducibility.
For clinical application, it is not only necessary to determine measurement variability, but also to assess reliability. Reliability is defined as ‘the degree to which measurement is free from measurement error’ (8). Reliability in our datasets was good (ICC > 0.8; Chapter 5 and 6). It is important to realize that reliability depends on the range in measurements (for example tumor uptake) and this is not directly generalizable between different datasets.
10
Summary and discussion
201