Program Information
BEST IN PHYSICS (JOINT IMAGING-THERAPY): Using FDG PET and CT Radiomics Features to Predict FMISO Uptake in Head and Neck Cancer
M Crispin Ortuzar*, A Apte, M Grkovski, J H Oh, N Y Lee, J L Humm, J O Deasy, Memorial Sloan-Kettering Cancer Center, New York, NY
Presentations
WE-G-605-2 (Wednesday, August 2, 2017) 4:30 PM - 6:00 PM Room: 605
Purpose: The purpose of this study is to assess whether a radiomics signature built from FDG PET and CT features could predict the maximum ¹⁸F-fluoromisonidazole (FMISO) uptake per lesion in head and neck cancer.
Methods: The analysis included 100 head and neck cancer patients undergoing chemoradiotherapy treatment at a single institution between 2011 and 2016. All of them received pre-treatment FDG and FMISO PET/CT. 134 lesions with a volume larger than 10 cc and delineated by physicians for treatment planning were analyzed, holding out 46 as a validation set. The level of hypoxia was defined as the maximum tumor-to-blood ratio (TBRᵐᵃˣ) of FMISO uptake measured at ~150 minutes post injection. Radiomics features were extracted from FDG PET (first order and run length features only) and CT (all orders). Lasso regression was performed on cross validation to reduce the number of features. The final model was found by running lasso again on the whole training set, using the reduced feature collection and including interaction terms. Performance was assessed by means of the mean square error (MSE) and area under the curve (AUC) analysis, assuming a hypoxia threshold of TBRᵐᵃˣ=1.4. Two models were developed: one using only FDG PET features, and another one using both FDG PET and CT.
Results: The mean FMISO TBRᵐᵃˣ of the population was 1.82 (SD = 0.65). The model built using only FDG PET features had an AUC of 0.76 and MSE of 0.27 on the validation dataset. The model built using all features had an AUC of 0.73 and an MSE of 0.29 on the validation dataset.
Conclusion: FDG PET features can be used to predict FMISO TBRᵐᵃˣ in head and neck cancer. Further analysis is needed to assess the complementarity of PET and CT features. Eventually, a validation study using multi-institutional datasets will be required.
Funding Support, Disclosures, and Conflict of Interest: This work was partially supported by NIH grant #1 R01 CA157770-01A1.
Contact Email: