7th ICHNO Abstract book

7th ICHNO 7 th ICHNO Conference International Conference on innovative approaches in Head and Neck Oncology 14 – 16 March 2019 Barcelona, Spain __________________________________________________________________________________________ page 59

evaluation were classification accuracy and area under the curve (AUC). Results The best performance was obtained from ADC-based models (85.7% using 55 features, AUC 0.85). T1WI and T2WI gave worse results (peak accuracy around 60%, AUC Radiomic analysis of pre-treatment imaging provides a potential, non-invasive, way to predict response to IC. Radiomics based on T1WI and T2WI was not useful in predicting response to IC, while predictive models based on ADC features reached up to 85% accuracy and AUC of 0.85. PO-116 Diagnostic and prognostic impact of 18F-FDG PET/CT imaging in patients with nasopharyngeal carcinoma J. Dura Esteve 1 , M. Alarza Cano 1 , A. Ruiz Alonso 1 , L. Iglesias Docampo 2 , A.C. Hernández Martínez 3 , A. Mera Errasti 1 , I. Alda Bravo 1 , M. Fasano Moore 1 , J.F. Pérez- Regadera Gómez 1 1 Hospital Universitario 12 de octubre, Radiation Oncology, Madrid, Spain; 2 Hospital Universitario 12 de octubre, Medical Oncology, Madrid, Spain; 3 Hospital Universitario 12 de Octubre, Nuclear Medicine, Madrid, Spain Purpose or Objective Describing the characteristics of patients diagnosed with nasopharyngeal carcinoma (NPC). To determine whether primary lesion maximum standardized uptake values (SUV max ) in 18 F-FDG PET/CT can be related to tumor responses after radiation therapy in NPC. Material and Methods This study retrospectively analyzed 38 patients with pathologically proven NPC, which received radiation therapy as main treatment. They all underwent 18 F-FDG PET/CT as staging and radiotherapy planning system. Results The mean age at diagnosis was 51 years, and 84% of patients were men. Lymphoepithelioma-like was the most frequent histology (61%), followed by squamous cell (22%) and undifferentiated (16%). Epstein-Barr virus DNA was found in 89% of the histological biopsies. The most prevalent stage was stage III (29%), followed by IV-A (26%), and stage II and I (21% and 12%). No patients with stage IV- C were included. In patients who underwent previous CT, 18 F-FDG PET/CT modified the final stage in 36% of them. Of those cases, in 55% the 18 F-FDG PET/CT staging was more advanced than obtained with CT. Mean SUV max observed was 19,80 (7,10- 48.30 CI 95%). The RT technique used was IMRT in 78% and 3D in 22% of cases. Concurrent chemotherapy was administered in 88% of patients. Clinical tumor response was assessed with 18 F-FDG PET/CT at 12-14 weeks after radiation therapy in a total of 16 patients. Of those, 56% accomplished a complete response and 31% showed a partial response. Progressive disease was found on 15% of cases. The Chi-square test was used to determine whether increased SUV max values were related to tumor responses with a p-value of 0.09. Conclusion 18 F-FDG PET/CT is a test with a high sensitivity range for NPC and may modify a significant proportion of CT staging. Moreover, this has a relevant implication in modifying the radiation therapy delineation. Thus, 18 F-FDG PET/CT around 0.6). Conclusion

At the time of this study there were no published consensus guidelines for CTVp delineation. This resulted in the use of different expansion margins and poor agreement which will hopefully improve with implementation of recently published guidelines. Although nearly all participants used identical guidelines for CTVe there were large discrepancies in neck levels selected and volumes delineated. This shows that additional teaching in target volume delineation is necessary as this paper demonstrates that availability and implementation of guidelines alone is not enough to guarantee uniform delineation. PO-115 Radiomics-based prediction of response to induction chemotherapy in sinonasal cancer M. Bologna 1 , G. Calareso 2 , C. Resteghini 3 , S. Sdao 2 , E. Montin 4 , V. Corino 1 , L. Mainardi 1 , L. Licitra 5 , P. Bossi 3 1 Politecnico di Milano, Department of Electronics- Information and Bioengineering DEIB, Milano, Italy; 2 Fondazione IRCCS Istituto Nazionale dei Tumori, Department of Radiology, Milano, Italy; 3 Fondazione IRCCS Istituto Nazionale dei Tumori, Department of Head and Neck Medical Oncology, Milano, Italy; 4 New York University School of Medicine, Department of Radiology, New York, USA; 5 University of Milan, Department of Oncology and Emato-Oncology, Milano, Italy Purpose or Objective To develop models for prediction of response to induction chemotherapy (IC) using MRI-based radiomic features in sinonasal cancers. Material and Methods Forty-two patients with sinonasal cancer (53 ± 12 years, 34 men) were considered in this study. Different types of MRIs were acquired at baseline: T1-weighted images (T1WI), T2-weighted (T2WI) images and diffusion weighted images (DWI). In particular, 41 patients had T1WI, 42 patients had T2WI and 28 patients had DWI. MRI scanners with 1.5T were used. The T1WI and T2WI were acquired using spin-echo sequences, while echo-planar imaging was used for the acquisition of the DWI.Clinical data were also acquired for each patient. In particular, the response to IC was evaluated using the Response and Evaluation Criteria in Solid Tumors (RECIST). Using the RECIST score, patients were divided in responders (partial or complete responders) and non-responders (stable or progressive disease). The distribution of the two classes was even (21 responders and 21 non-responders). The tumor region was manually segmented by an expert radiologist. Intensity standardization was applied to T1WI and T2WI to normalize the range of intensities of the different images. In particular, quantile 0 to 0.98 of the intensities of the images were linearly mapped to the range 0-5000. DWI were used to compute the Apparent Diffusion Coefficient (ADC), which is a quantitative measure and therefore does not need normalization. For each image type (T1WI, T2WI and ADC) a separate radiomic analysis was performed. In particular, 89 radiomic features divided in 3 categories (first-order statistics, textural and geometrical features) were extracted. Z-score normalization was performed to normalize the ranges of variability of the features. Several predictive models were tested, differing for type of image analyzed (T1WI, T2WI or ADC), and number of features (added to the model according to a rank based on statistical significance). Logistic regression was used as classification algorithm. Leave-one-out cross-validation was performed to evaluate each model. The metrics of