Abstract Book

S224

ESTRO 37

2 Cancer Hospital of Shantou University Medical College, Radiation Oncology, Shantou, China Purpose or Objective This study aims to explore the prognostic value of computed tomography (CT) image biomarkers (IBMs) in combination with clinical parameters for local recurrence (LR), regional recurrence (RR), distant metastasis (DM) and disease free survival (DFS) in head and neck cancer (HNC) patients. Material and Methods This is a retrospective analysis in a prospective cohort of consecutive HNC patients treated between July 2007 and June 2016 at our department. IBMs representing geometric, CT intensity and textural characteristics of the primary tumor (PT) and positive lymph nodes (LN), which were extracted from the CT images with contrast enhancement. Multivariable Cox regression analysis was performed to identify potential prognostic factors for LR, RR, DM and DFS. These prognostic factors were internally validated by bootstrapping. The concordance-index (c- index) was determined to assess the models' discriminative power. An overview of IBMs extraction process, analysis and future application is shown in Figure 1.

(Table1).

Conclusion The addition of IBMs from the primary tumor and positive lymph nodes improved the predictive performance of prediction models for LR, RR, DM and DFS, compared to models containing clinical and dose factors only, in HNC patients. PV-0427 CBCT based estimation of delivered dose is not more predictive for NTCP than planned dose C. Hvid 1 , U. Elstrøm 2 , K. Jensen 1 , C. Grau 1 1 Aarhus University Hospital, Dept of Oncology, Aarhus C, Denmark 2 Aarhus University Hospital, Dept of Medical Physics, Aarhus C, Denmark Purpose or Objective Head and neck radiation therapy can lead to significant late morbidity, including dysphagia and xerostomia. The risk of developing morbidity depends on the dose delivered to the organs at risk (OAR). Anatomical changes can occur during treatment, which may cause deviations between planned and actually delivered dose to OAR. The purpose of this study was to investigate if cone-beam computed tomography (CBCT) based determination of anatomical changes and delivered dose in organs at risk improved normal tissue complication modelling compared to using planned dose only. Material and Methods The study included 114 patients treated with intensity modulated radiation therapy (IMRT) for head and neck squamous cell carcinoma. Parotid glands and pharyngeal constrictor muscles (PCMs) were delineated on planning CT and CBCTs. Delivered dose was estimated by recalculating and summing dose on weekly CBCTs. Late morbidity data for xerostomia and dysphagia were collected from follow-up visits and dichotomized with physician scored moderate to severe outcomes as endpoint. Univariate logistic regression modeling was used to generate dose response curves for delivered and planned dose, which were compared by their γ 50 and D 50 During treatment parotid glands shrank by 4.05 ml (18.4%) p<0.01 and CBCT based delivered dose was 0.54 Gy (2%) p<0.01 higher than planned dose. No significant changes in volume and delivered dose to PCMs were found. Figure 1 shows volumetric (black line) and dosimetric (grey line) relative changes from baseline for parotid glands and PCMs. Dose response curves based on delivered doses were not significantly different from curves based on planned doses. Figure 2 shows dose response curves for contralateral parotid gland and superior PCM. values. Results

Results A total of 472 patients were analyzed. Overall 84 (18%), 44 (9%), 50 (11%) and 198 (42%) events occurred for LR, RR, DM and DFS, respectively. In the clinical multivariable analysis, WHO performance status (WHO PS) were significantly associated with LR, while N stage, WHO PS, gender and D95% of GTV were associated with RR, N stage and Dmean of GTV with DM and N stage, WHO PS and age with DFS. Multivariable analysis including IBMs identified volume of bounding box (VB) as independent prognostic factor for LR, energy and skewness of LN for RR, volume density and autocorrelation of gray level co- occurrence matrix for DM, and energy and VB for DFS. The VB refers to the volume of the smallest cube containing the tumor. The volume density is tumor volume divided by the VB. Energy and skewness are features representing the first order statistical property of the image, describing the overall density of the tumor and the asymmetry of the probability distribution of gray levels of each voxels about its mean. The autocorrelation describes the linear dependency of gray level on those of neighbouring voxels, indicating the tissue heterogeneity. A lower autocorrelation value of an image indicates a larger heterogeneity of the image intensity, and associated with worse metastasis free survival. Adding these IBMs to the prediction models consisting of clinical and dose parameters only, model performance in terms of c-indexes for LR, RR, DM and DFS significantly improved to 0.66, 0.84, 0.75 and 0.70, respectively