Click here to


Are you sure ?

Yes, do it No, cancel

Intra-Treatment 18F-FDG PET/CT Radiomic Signature Predicts In-Field Recurrence Following Definitive Chemo-Radiation Therapy for Oropharyngeal Cancer

K Lafata*, Y Chang , C Wang , Y Mowery , D Brizel , F Yin , Duke University Medical Center, Durham, NC


(Wednesday, 7/17/2019) 7:30 AM - 9:30 AM

Room: 225BCD

Purpose: To investigate the association between oropharyngeal cancer (OPC) recurrence after treatment and radiomic features derived from both pre-treatment and intra-treatment PET/CT images. We hypothesized that intra-treatment radiomics data would provide better prognostic value compared to baseline pre-treatment radiomics data

Methods: Sixty patients undergoing definitive (70Gy) chemo-radiation for OPC were enrolled in an IRB-approved prospective study. 18F-FDG PET/CT images were acquired both pre-treatment and intra-treatment (@20Gy) on a single Siemens Biograph mCT PET-CT scanner. The gross tumor volume was segmented on both image sets, from which 61 radiomic features were extracted as potential biomarkers for treatment response. These quantitative features collectively captured tumor morphology, intensity, fine-texture, and coarse-texture. The primary clinical end-point was in-field cancer recurrence, which was assessed by diagnostic PET/CT at 12 weeks post-treatment, serial physical exam, and biopsy. LASSO regularized logistic regression modeling was implemented to quantify the multivariate relationship between the radiomic features and cancer recurrence. Separate models were developed in parallel, with identical hyper-parameterization, to compare the relative performance between the pre-treatment and intra-treatment radiomics data. Model performance was based on Receiver Operating Characteristic (ROC) curve analysis, and generalization was evaluated using stratified Monte Carlo cross-validation (200 iterations, 80% training, 20% testing, equal event ratios).

Results: The 20 Gy intra-treatment PET/CT radiomic model (AUC=0.70±0.03) was more predictive of cancer recurrence than the analogous pre-treatment model (AUC=0.54±0.02, p<0.001). The most important radiomic features selected from CT and PET were Small-Size-High-Gray-Level-Emphasis and Sum Average, respectively. These quantitative results suggest that dense tumors with a heterogeneous fine texture and high FDG-uptake were associated with an increased risk of recurrence.

Conclusion: Intra-treatment PET/CT radiomics data may provide an early indication of treatment failure. This may facilitate adaptive treatment strategies including modification of treatment volumes and novel dose escalation techniques, and closer follow-up for patients at high-risk of in-field recurrence.


CT, FDG PET, Radiation Therapy


IM/TH- Image Analysis (Single modality or Multi-modality): Computer-aided decision support systems (detection, diagnosis, risk prediction, staging, treatment response assessment/monitoring, prognosis prediction)

Contact Email