Abstract
353
Objectives: Conventional quantitative imaging metrics, such as the standardized uptake value (SUV) or the apparent diffusion coefficient (ADC), are typically measured and reported in a summarized fashion for a tumor (e.g., mean ADC, SUV max, or SUV peak). As a result, these metrics may fail to capture important information about treatment effects, especially when changes are heterogeneous. In such scenarios, voxelwise analysis, which retains voxel-level data, has the potential to improve the accuracy of tumor response assessments. However, the intrinsic variability of voxel-level data is largely unknown. Thus, the aim of our study was to determine the voxelwise test-retest repeatability of [18F]FDG-PET/MRI metrics, in a population of patients with malignant solid tumors of the pelvis.
Methods: This prospective, IRB-approved study enrolled subjects with pathology-proven pelvic malignancies that were either treatment-naïve or had not been treated for at least 30 days. Subjects, who were at least 18 years of age and provided written informed consent, underwent [18F]FDG-PET/MRI on two separate occasions separated by 1-7 days. A 10 mCi dose of [18F]FDG was administered on both occasions. Utilizing a single PET/MR scanner, each imaging session included a 30 min list mode PET acquisition (single station, centered craniocaudally on the tumor) and diffusion-weighted imaging, in addition to standard MR sequences. A 3 min interval of PET data was selected for static reconstruction from each imaging session to achieve identical effective uptake times for the test and retest scans. The pelvic tumor was manually contoured on the static PET images and ADC maps for both imaging sessions, resulting in a total of 4 volumes of interest (VOIs) for each subject. A deformable co-registration algorithm was run to achieve identical tumor contours (with identical constituent voxel sizes) for all 4 VOIs. SUVs, ADCs, and SUV/ADC ratios were extracted for each voxel in the contour for both time points. The coefficient of determination (R2) and repeatability coefficient (RC) were determined on a voxelwise basis for each subject.
Results: 11 subjects completed the necessary imaging for repeatability analysis. The mean age was 48.7 years. 10 of 11 subjects had new diagnoses of cervical cancer. The median [range] uptake time was 96 min [88-105 min]. The median [range] number of voxels assessed per subject was 65,332 [13,874-451,539]. Across all subjects, median [range] voxelwise R2 value was 0.77 [0.64-0.94] for SUV, 0.37 [0.22-0.78] for ADC, and 0.76 [0.57-0.98] for SUV/ADC. Across all subjects, the median [range] voxelwise RC value was 2.4 [1.8-4.9] for SUV, 237 x 10-6 mm/s2 [157-367 x 10-6 mm/s2] for ADC, and 3.3 x 103 s2/mm [2.5-6.4 x 103 s2/mm] for SUV/ADC.
Conclusions: Voxelwise correlations between imaging sessions are strong for SUV and SUV/ADC but relatively poor for ADC. For 95% of voxels, the SUV, ADC, and SUV/ADC of an individual voxel is expected to fall within ± 2.4, ± 237 x 10-6 mm/s2, and ± 3.3 x 103 s2/mm, respectively, of its initial value on repeat measurement. Voxelwise quantitative imaging metrics are promising biomarkers for the treatment response assessment of pelvic solid tumors. Acknowledgements: This project was supported by NIH grant U01 CA140204.