Voxel-wise Repeatability of [18F]FDG PET/MRI Quantitative Metrics in Cervical Cancer

Background Unlike whole-tumor summary metrics, voxel-wise metrics in positron emission tomography (PET) and magnetic resonance imaging (MRI) capture tumor heterogeneity. This study aimed to determine the voxel-wise repeatability of the standardized uptake value (SUV), apparent diffusion coefficient (ADC), and SUV/ADC ratio, compared with whole-tumor summary metrics, on 2-deoxy-2-[ 18 F]fluoro-D-glucose ([ 18 F]FDG) PET/MRI. Methods Ten patients with newly diagnosed cervical cancer underwent two [ 18 F]FDG-PET/MRIs 1–7 days apart. In MIM version 6.7, cervical tumor volumes of interest (VOIs) were manually delineated on PET images and ADC maps for both sessions. A deformable registration algorithm was run, using either the first session PET (PET1) or ADC (ADC1) VOI as the deformation target for the other three VOIs. SUV, ADC, and SUV/ADC values were extracted for each voxel. Intraclass correlation coefficients (ICCs) and Bland-Altman statistics were calculated. Statistical significance was assessed via the Wilcoxon signed rank test. Results SUV had better voxel-wise relative repeatability coefficients (RC-rel) for the ADC1 versus PET1 target (50.2% vs. 56.7% [ P = 0.04]). Similarly, ADC had better voxel-wise RC-rel for the PET1 versus ADC1 target (30.9% vs. 36.6% [ P = 0.01]). SUV/ADC had no significant target-related difference. For SUV, subject-level voxel-wise ICC interquartile ranges (0.72–0.91) were similar to cohort-level ICCs for SUV mean (0.81–0.83). In contrast, for ADC, subject-level ICC interquartile ranges (0.48–0.72) were substantially lower than cohort-level ICCs for ADC mean (0.84–0.86). Conclusion On [ 18 F]FDG-PET/MRI, deformation targets have small but significant effects on voxel-wise repeatability. For SUV and ADC, whole-tumor metrics had better repeatability than voxel-wise metrics.