Page 113 - 18F-FDG PET as biomarker in aggressive lymphoma; technical and clinical validation
P. 113
Workflow optimization of MTV in DLBCL Table 3. Most preferred method* per observer for Workflow A.
Patient Observer 1
1 41%MAX
2 41%MAX
3 A50%P
4 SUV≥4.0
5 SUV≥4.0
6 SUV≥4.0
7 A50%P
8 A50%P
9 A50%P
10 A50%P
11 SUV≥4.0
12 41%MAX/SUV≥4.0
Observer 2
41%MAX 41%MAX/A50%P/SUV≥4.0 41%MAX
A50%P
A50%P
41%MAX/A50%P
A50%P
41%MAX/A50%P 41%MAX
A50%P
41%MAX
41%MAX
Observer 3
SUV≥4.0 SUV≥2.5 SUV≥4.0 SUV≥4.0 SUV≥4.0 SUV≥4.0 SUV≥2.5 SUV≥4.0 SUV≥4.0 A50%P A50%P A50%P
*Each observer indicated their “preferred segmentation” for individual lesions. The most preferred method per patient was defined as the method most often noted as “preferred segmentation”.
Figure 2 shows the modified MTVs approved by a nuclear medicine physician (final MTV ). Figure 3 shows a scatterplot of the correlation between the preselected and final MTV in Workflow C. Interestingly, the same outlier (patient 10) occurred as in Fig. 1, but contrary to this, two observers now decided to keep the entire liver in the preselection of Workflow C while they removed the liver uptake in Workflow B. For the final MTV, observer 2 had to remove the liver uptake after the check by the experienced NM physician. In another patient (patient 11), the preselection missed many small bone lesions, which were added manually. Figure 4 shows the Bland-Altman plot of the preselected and final MTV in Workflow C. The 95 % limits of agreement ranged widely (− 525 to 458). The differences between preselected and final MTV did not have a normal distribution according to the SW test (P = 0.002). After excluding patients 10 and 11 (Figs. 3 and 4) described as outliers, the mean difference had a normal distribution (P = 0.106). The plot shows both the original—as well as the recalculated 95 % limits of agreement after exclusion of the outliers.
111
5
