A positron emission tomography radiomic signature for distant metastases risk in oropharyngeal cancer patients treated with definitive chemoradiotherapy.

Abstract

Disease recurrence and distant metastases (DM) are major concerns for oropharyngeal cancer (OPC) patients receiving definitive chemo-radiotherapy. Here, we investigated whether pre-treatment primary tumor positron emission tomography (PET) features could predict progression-free survival (PFS) or DM. Primary tumors were delineated on pre-treatment PET scans for patients treated between 2005 and 2018 using gradient-based segmentation. Radiomic image features were extracted, along with SUV metrics. Features with zero variance and strong correlation to tumor volume, stage, p16 status, age or smoking were excluded. A random forest model was used to identify features associated with PFS. Kaplan-Meier methods, Cox regression and logistic regression with receiver operating characteristics (ROC) and 5-fold cross-validated areas-under-the-curve (CV-AUCs) were used. A total of 114 patients were included. With median follow-up 40months (range: 3-138months), 14 patients had local recurrence, 21 had DM and 38 died. Two-year actuarial local control, distant control, PFS and overall survival was 89%, 84%, 70% and 84%, respectively. The wavelet_LHL_GLDZM_LILDE feature slightly improved PFS prediction compared to clinical features alone (CV-AUC 0.73 vs. 0.71). Age>65years (HR=2.64 (95%CI: 1.36-5.2), p=0.004) and p16-negative disease (HR=3.38 (95%CI: 1.72-6.66), p<0.001) were associated with poor PFS. A binary radiomic classifier strongly predicted DM with multivariable HR=3.27 (95%CI: 1.15-9.31), p=0.027, specifically for patients with p16-negative disease with 2-year DM-free survival 83% for low-risk vs. 38% for high-risk patients (p=0.004). A radiomics signature strongly associated with DM risk could provide a tool for improved risk stratification, potentially adding adjuvant immunotherapy for high-risk patients.