Page 133 - 89Zr-Immuno-PET:Towards a Clinical Tool to Guide Antibody-based Therapy in Cancer

Interobserver reproducibility of 89Zr-immuno-PET
tumor lesions on immuno-PET were still not fully clear when viewing both the immuno-PET and the 18F-FDG-PET. After viewing the corresponding 18F-FDG- PET, 25% of the VOI were adapted by at least 1 observer (Table 2).
Semi-automatically generated VOI were accepted by all three observers in 45% of all VOI (Table 2). Inclusion of background was the main reason for failure of semi-automatic VOI.
Table 2 Effect of viewing the 18F-FDG-PET on manual tumor delineation on immuno-PET and success rate of semi-automatic delineation
89Zr-rituximab 89Zr-cetuximab 89Zr-trastuzumab All D3 D6 D3 D6 D4
% of VOI changed by ≥ 1 observer after
viewing the 18F-FDG-PET 19% 34% 0% 10% 31%* 25%
% of VOI accepted after semi-automatic 58% 66% 14% 30% 21% 45% delineation
*No 18F-FDG-PET available for patient 17
Eligibility criteria for VOI delineation 6 Measurement variability for SUVmax was > 0 in 25% (26/103) of the manually
delineated VOI.
In 4% of the cases (4/103) a different structure (e.g. another tumor lesion) was delineated by at least 1 observer (2/32 (D6) for 89Zr-rituximab; 1/10 (D6) for 89Zr-cetuximab and 1/28 (D4) for 89Zr-trastuzumab). In 15% of the cases (15/103), the voxel with the maximum intensity was located at the border of the VOI (3/26 (D3) and 2/30 (D6) for 89Zr-rituximab; 5/7 (D3) and 1/9 (D6) for 89Zr-cetuximab and 4/27 (D4) for 89Zr-trastuzumab).
Application of eligibility criteria resulted in exclusion of 19 VOI, as tumor contrast was apparently insufficient for correct VOI delineation.
Interobserver reproducibility
Interobserver variability.
Relative interobserver variability (CoV) was not correlated with tumor uptake (SUVmean) (Figure 3). Therefore, interobserver variability is reported as a relative value per individual VOI and per datagroup (e.g. timepoint, mAb) (Table 3). For all VOI combined (n=103), interobserver variability was 0% (0-2) for SUVmax, 0% (0-12) for SUVpeak and 7% (5-14) for SUVmean. Manual delineation resulted in an interobserver variability of 35% (21-49) for delineated volume and 30% (17-44) for TLU.
131