Room: Stars at Night Ballroom 2-3
Purpose: Tumor voxel dose response can be assessed using serial FDG-PET/CT imaging feedback. This study will investigate the predictive capability of tumor voxel dose response using one or two imaging feedback.
Methods: Â¹â?¸18F-FDG PET/CT images obtained at pre and weekly during chemo-radiotherapy from 18 HNC patients. All weekly PET/CT images of each patient were registered voxel-by-voxel to their pre-treatment PET/CT image to construct a tumor voxel dose response matrix (DRM). For each tumor voxel v, the reference DRM (DRM_ref) was derived using the linear regression on logarithm of the first 4 weekly SUV ratios (rSUV), defined as ln rSUV = ln [SUVi(v)/SUV0(v)], i = 1 to 4, which represents the average change rate of tumor voxel metabolism with respect to the delivered treatment dose. In addition, the DRM was predicted using one PET image acquired at the treatment week 2 (DRM_pre(week 2)) or 3 (DRM_pre(week 3)), or two PET images acquired at the week 2 and 4 (DRM_pre(week 2&4)) respectively. ROC analysis was performed to test the predictive capability of each DRM, where the true positive/negative is defined as a tumor has a resistant sub-volume larger/smaller than a cutoff value would be a local-regional failure/control tumor. The resistant sub-volume, V(DRM(v) > 0.9), was applied in the test with respect to the clinical followup of tumor local-regional failure/control.
Results: The DRM_pre(week 2), DRM_pre(week 3) and DRM_pre(week 2&4) achieved the mean (SD) Spearman correlation coefficient of 0.87 (0.07), 0.9 (0.06) and 0.97 (0.04) to the DRM_ref. The predictive values of V(DRM(v) > 0.9) calculated from the DRM_pre(week 2), DRM_pre(week 3), DRM_pre(week 2&4) and DRM_ref were comparable with the (AUC)s of 0.95, 0.96, 0.83 and 0.98 respectively.
Conclusion: The tumor voxel dose response matrix DRM constructed using one or two FDG-PET/CT imaging feedback has good predictive capability to assess local radioresistance in tumor.