No abstract

No abstract

Background and purpose: In head and neck squamous cell carcinoma (HNSCC) (chemo)radiotherapy is increasingly used to preserve organ functionality. The purpose of this study was to identify predictive pretreatment DWI-and 18 F-FDG-PET/CT-parameters for treatment failure (TF), locoregional recurrence (LR) and death in HNSCC patients treated by (chemo)radiotherapy. Materials and methods: We retrospectively included 134 histologically proven HNSCC patients treated with (chemo)radiotherapy between 2012-2017. In 58 patients pre-treatment DWI and 18 F-FDG-PET/CT were performed, in 31 patients DWI only and in 45 patients 18 F-FDG-PET/CT only. Primary tumor (PT) and largest lymph node (LN) metastasis were quantitatively assessed for TF, LR and death. Multivariate analysis was performed for 18 F-FDG-PET/CT and DWI separately and thereafter combined. In patients with both imaging modalities, positive and negative predictive value in TF and differences in LR and death, were assessed. Results: Mean follow-up was 25.6 months (interquartile-range; 14.0-37.1 months). Predictors of treatment failure, corrected for TNM-stage and HPV-status, were SUV max-PT , ADC max-PT , total lesion glycolysis (TLG-LN), ADCp20-LN (P = 0.049, P = 0.024, P = 0.031, P = 0.047, respectively). TLG-PT was predictive for LR (P = 0.003). Metabolic active tumor volume (MATV-PT) (P = 0.003), ADC GTV-PT (P < 0.001), ADCSD (P = 0.048) were significant predictors for death. In patients with both imaging modalities SUV max-PT remained predictive for treatment failure (P = 0.049), TLG-LN for LR (P = 0.003) and ADC GTV-PT for death (P < 0.001). Higher predictive value for treatment failure was found for the combination of SUV max-PT and ADC max-PT , compared to either one separately. Conclusion: Both DWI-and 18 F-FDG-PET/CT-parameters appear to have predictive value for treatment failure, locoregional recurrence and death. Combining SUV max-PT and ADC max-PT resulted in better prediction of treatment failure compared to single parameter assessment.

A major challenge in radiomics is assembling data from multiple centers. Sharing data between hospitals is restricted by legal and ethical regulations. Distributed learning is a technique, enabling training models on multicenter data without data leaving the hospitals ("privacy-preserving" distributed learning). This study tested feasibility of distributed learning of radiomics data for prediction of two year overall survival and HPV status in head and neck cancer (HNC) patients. Pretreatment CT images were collected from 1174 HNC patients in 6 different cohorts. 981 radiomic features were extracted using Z-Rad software implementation. Hierarchical clustering was performed to preselect features. Classification was done using logistic regression. In the validation dataset, the receiver operating characteristics (Roc) were compared between the models trained in the centralized and distributed manner. No difference in ROC was observed with respect to feature selection. The logistic regression coefficients were identical between the methods (absolute difference <10 −7). In comparison of the full workflow (feature selection and classification), no significant difference in ROC was found between centralized and distributed models for both studied endpoints (DeLong p > 0.05). In conclusion, both feature selection and classification are feasible in a distributed manner using radiomics data, which opens new possibility for training more reliable radiomics models.