Background: Radiomics is a promising methodology for quantitative analysis and description of radiological images using advanced mathematics and statistics. Tumor delineation, which is still often done manually, is an essential step in radiomics, however, inter-observer variability is a well-known uncertainty in radiation oncology. This study investigated the impact of inter-observer variability (IOV) in manual tumor delineation on the reliability of radiomic features (RF). Methods: Three different tumor types (head and neck squamous cell carcinoma (HNSCC), malignant pleural mesothelioma (MPM) and non-small cell lung cancer (NSCLC)) were included. For each site, eleven individual tumors were contoured on CT scans by three experienced radiation oncologists. Dice coefficients (DC) were calculated for quantification of delineation variability. RF were calculated with an in-house developed software implementation, which comprises 1404 features: shape (n = 18), histogram (n = 17), texture (n = 137) and wavelet (n = 1232). The IOV of RF was studied using the intraclass correlation coefficient (ICC). An ICC >0.8 indicates a good reproducibility. For the stable RF, an average linkage hierarchical clustering was performed to identify classes of uncorrelated features. Results: Median DC was high for NSCLC (0.86, range 0.57–0.90) and HNSCC (0.72, 0.21–0.89), whereas it was low for MPM (0.26, 0–0.9) indicating substantial IOV. Stability rate of RF correlated with DC and depended on tumor site, showing a high stability in NSCLC (90% of total parameters), acceptable stability in HNSCC (59% of total parameters) and low stability in MPM (36% of total parameters). Shape features showed the weakest stability across all tumor types. Hierarchical clustering revealed 14 groups of correlated and stable features for NSCLC and 6 groups for both HNSCC and MPM. Conclusion: Inter-observer delineation variability has a relevant influence on radiomics analysis and is strongly influenced by tumor type. This leads to a reduced number of suitable imaging features.

Purpose: An association between radiomic features extracted from CT and local tumor control in the head and neck squamous cell carcinoma (HNSCC) has been shown. This study investigated the value of pretreatment functional imaging (18F-FDG PET) radiomics for modeling of local tumor control. Material and Methods: Data from HNSCC patients (n = 121) treated with definitive radiochemotherapy were used for model training. In total, 569 radiomic features were extracted from both contrast-enhanced CT and 18F-FDG PET images in the primary tumor region. CT, PET and combined PET/CT radiomic models to assess local tumor control were trained separately. Five feature selection and three classification methods were implemented. The performance of the models was quantified using concordance index (CI) in 5-fold cross validation in the training cohort. The best models, per image modality, were compared and verified in the independent validation cohort (n = 51). The difference in CI was investigated using bootstrapping. Additionally, the observed and radiomics-based estimated probabilities of local tumor control were compared between two risk groups. Results: The feature selection using principal component analysis and the classification based on the multivariabale Cox regression with backward selection of the variables resulted in the best models for all image modalities (CI_CT = 0.72, CI_PET = 0.74, CI_PET/CT = 0.77). Tumors more homogenous in CT density (decreased GLSZM_{size_zone_entropy}) and with a focused region of high FDG uptake (higher GLSZM_SZLGE) indicated better prognosis. No significant difference in the performance of the models in the validation cohort was observed (CI_CT = 0.73, CI_PET = 0.71, CI_PET/CT = 0.73). However, the CT radiomics-based model overestimated the probability of tumor control in the poor prognostic group (predicted = 68%, observed = 56%). Conclusions: Both CT and PET radiomics showed equally good discriminative power for local tumor control modeling in HNSCC. However, CT-based predictions overestimated the local control rate in the poor prognostic validation cohort, and thus, we recommend to base the local control modeling on the 18F-FDG PET.

Objectives: Human papillomavirus (HPV) positive oropharyngeal cancer (oropharyngeal squamous cell carcinoma, OPSCC) is biologically and clinically different from HPV negative OPSCC. Here, we evaluate the use of a radiomic approach to identify the HPV status of OPSCC. Methods: Four independent cohorts, totaling 778 OPSCC patients with HPV determined by p16 were collected. We randomly assigned 80% of all data for model training (N = 628) and 20% for validation (N = 150). On the pre-treatment CT images, 902 radiomic features were calculated from the gross tumor volume. Multivariable modeling was performed using least absolute shrinkage and selection operator. To assess the impact of CT artifacts in predicting HPV (p16), a model was developed on all training data (Mall) and on the artifact-free subset of training data (Mno art). Models were validated on all validation data (Vall), and the subgroups with (Vart) and without (Vno art) artifacts. Kaplan–Meier survival analysis was performed to compare HPV status based on p16 and radiomic model predictions. Results: The area under the receiver operator curve for Mall and Mno art ranged between 0.70 and 0.80 and was not significantly different for all validation data sets. There was a consistent and significant split between survival curves with HPV status determined by p16 [p = 0.007; hazard ratio (HR): 0.46], Mall (p = 0.036; HR: 0.55) and Mno art (p = 0.027; HR: 0.49). Conclusion: This study provides proof of concept that molecular information can be derived from standard medical images and shows potential for radiomics as imaging biomarker of HPV status. Advances in knowledge: Radiomics has the potential to identify clinically relevant molecular phenotypes.