Click here to


Are you sure ?

Yes, do it No, cancel

Prediction of Local Persistence/Recurrence After Radiation Therapy Treatment of Head and Neck Cancer From PET/CT Using a Multi-Objective Radiomics Model

Q Zhang*, Z Zhou , G Qin , P Li , J Shah , N Pham , S Gottumukkala , Z Moore , D Sher , J Wang , S Jiang , UT Southwestern Medical Center, Dallas, TX


(Sunday, 7/29/2018) 3:00 PM - 6:00 PM

Room: Exhibit Hall

Purpose: PET/CT is commonly utilized as post-radiation treatment surveillance for detection of cancer persistence/recurrence in head and neck cancer. Radiotherapy-induced inflammation may confound accuracy of residue disease detection, leading to false clinical decisions. The purpose of this study is to use a multi-objective radiomics model for predicting post-treatment local persistence/recurrence on PET/CT scans.

Methods: A total of 100 head and neck cancer patients with definitive radiation therapy at UT Southwestern Medical Center were retrospectively selected for the study. Patients of oropharynx, nasopharynx, hypopharynx, and larynx cancers were included. The median time from treatment completion to PET/CT imaging was 112 days (range: 89 days to 123 days). Local control was extracted, with 40 patients being classified as persistence/recurrence group (P/R), and 60 as non-persistence/recurrence group (NP/NR). Post-treatment PET/CT images with FDG were reviewed and contoured. Two hundred and fifty-seven image features including intensity features, textural features and geometric features were extracted for training a multi-objective radiomics model, in which both sensitivity and specificity were used as objectives during model training and feature selection. Support vector machine (SVM) with radial basis function kernel was used as the predictive model.

Results: In the predictive model, three imaging modalities (PET, CT, and their combination) were used as the input. Accuracy, area under the receiver operating characteristic curve (AUC), sensitivity, and specificity were taken as the evaluation criteria (see Table 1).The combination of PET and CT imaging in the multi-objective radiomics model shows a prediction accuracy of 80.0%, AUC of 72.7%, sensitivity of 70.0% and specificity of 86.6%.

Conclusion: This study demonstrates the feasibility of using multi-objective radiomics model from PET/CT images to predict the outcome for H/N cancer patients after radiation therapy. This prediction models allows for early detection of persistent/recurrent cancer after radiation therapy, and distinguish treatment-related inflammation from recurrence.


Computed Radiography, Diagnostic Radiology, Image Analysis


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