Optimal [18]F-Misonidazole PET threshold to locate SCC7 tumor hypoxia using EPR pO₂ as ground truth

Objectives: Tumor hypoxia is associated with resistance to therapy and tumor progression, and correlates negatively with patient survival [1, 2, 3]. ¹⁸F-Misonidazole (FMISO) is frequently used in clinical PET trials to measure and treat tumor hypoxia [4], but there is no universally accepted threshold to define tumor hypoxia with FMISO. This study uses electron paramagnetic resonance (EPR) pO₂ images as true hypoxia (pO₂ < 10 mmHg) [5] to calculate the optimal corresponding FMISO PET threshold for identifying hypoxic tumors in SCC7 tumor murine models of squamous cell carcinoma.

Methods: Imaging: Using SCC7 squamous cell carcinoma murine models (n=14), the tumor-bearing leg was immobilized in the plastic bed in a polysiloxane dental mold cast (GC America, Alsip, IL) with embedded fiducials to allow for co-registration between modalities. FMISO PET and EPR images were acquired in a hybrid PET/EPR system for simultaneous imaging [6], which gave the advantage of identical physiological conditions of the mouse. A tail-vein cannula was used to administer an oxygen-sensitive spin probe solution for EPR imaging. A bolus injection of ~230 uCi of FMISO (produced at the in-house cyclotron facility) was used for PET imaging; images were acquired 2-hours post-injection. T2-weighted images were acquired in a 9.4 Tesla small animal imager (Bruker, Erlangen, Germany) for registration and tumor/muscle contouring. Image analysis: Following MRI/EPR/PET registration in MATLAB, images were resampled to the PET image’s isotropic voxel resolution of [0.5 mm]³. The T2 MRI-based tumor and muscle contour were transformed to the PET and EPR images in units of tumor-to-muscle ratio (TMR) and pO₂, respectively. Using a custom-written script in MATLAB, ROC curves were generated for each tumor across all thresholds of PET TMR > 0 to 5.6 in increments of 0.2, using EPR pO₂ < 10 mmHg as true hypoxia. The accuracy (ACC) (fraction of true negatives and positives over all true/false negatives/positives), Dice Similarity Coefficient (DSC), and Hausdorff Distance (d_H) were used to quantify overlap between hypoxic regions as defined by EPR and PET. Because maximum ACC and DSC are both between 0 and 1, with 1 corresponding to highest overlap, d_H was normalized and subtracted from 1 (1 - || d_H ||) so that the highest value would also show maximum overlap. The peak mean of ACC, DSC, and 1 - ||d_H|| averaged over all tumors was used to determine the optimal PET threshold. The hypoxic fractions of tumor voxels based on resulting thresholds was also calculated to compare between modalities.

Results: For all tumors, the area under the ROC curve using pO₂ < 10 mmHg as gold standard was AUC = 0.739 (SE = 0.03). The peak ACC = 0.816 (SE = 0.02) corresponding to the PET threshold TMR > 2.4, and peak DSC = 0.485 (SE = 0.05) corresponding to a threshold TMR 2.0. At its minimum, mean d_H = 3.40 (SE = 0.2) mm at TMR > 2.4. The average value of ACC, DSC, and 1-|| d_H || showed a peak at TMR > 2.2 and pO₂ < 10 mmHg. The mean hypoxic fraction of EPR images was 0.20 (SE = 0.05), and of PET images was 0.19 (SE = 0.03), which was not significant based on the two-sample t-test (p = 0.51).

Conclusions: Based on this dataset of SCC7 squamous cell carcinoma murine models, the PET threshold of TMR > 2.2 has the highest ACC and DSC, and the lowest d_H, when compared to hypoxic tumor regions defined by EPR pO₂ < 10 mmHg. These results might help improve patient prognosis for more accurate hypoxia-based dose-painting treatment plans based on PET imaging. References: [1] Hockel, M et al. (1996). Cancer Res, 56(19): p. 4509-15.[2] Hockel, M and Vaupel, P (2001). J Natl Cancer Inst, 93(4): p. 266-76.[3] Brizel, DM et al. (1997). Int J Radiat Oncol Biol Phys, 38(2): p. 285-9.[4] Lopci, E et al. (2014). Am J Nucl Med Mol Imaging, 4: p. 365-384.[5] Epel, B et al. (2019). Int J Radiat Oncol Biol Phys, 103(4): p. 977-984.[6] Kim H et al. (2020). Nuclear Inst. and Methods in Physics Research Section, A, 959.

• Download figure
• Open in new tab
• Download powerpoint