ESTRO 36 Abstract Book

S228 ESTRO 36 _______________________________________________________________________________________________

by TME surgery and optional postoperative chemotherapy and an experimental arm B: short course 5 x 5 Gy radiation followed by six cycles of full-dose CAPOX or nine cycles of FOLFOX and TME surgery. Results A total of 920 patients were included between June 2011 and June 2016. At randomisation, 302 were cT4 and 828 were cN+, of whom 621 were considered cN2 disease and 137 as extramesorectal pelvic lymphnodes. Based on MRI, extramural vascular invasion was diagnosed in 275 patients, whereas the mesorectal fascia was threatened in 564 patients. Preliminary data show that median time between randomization and surgery was 15,9 weeks for arm A and 25,3 weeks for arm B. In arm B, 100% of the patients who started, completed the radiotherapy and 72% of patients completed all scheduled cycles of neoadjuvant chemotherapy after 5x5 Gy. Another 9% of patients completed the last course(s) without oxaliplatin. In arm A, 96% received all scheduled radiotherapy fractions and 94% of the patients received 5 weeks of preoperative capecitabine combined with radiotherapy. Open surgery was performed in 59% of the patients and 35% underwent an APR. In total, 19% of patients had a ypT0N0. For 4% of all patients a wait & watch strategy was applied. Of the operated patients, 89% had a negative circumferential resection margin (> 1 mm). Conclusion Compliance for neoadjuvant treatment was good in both treatment arms. Given the locally advanced state of most tumors, the ypT0N0 rate can be considered satisfactory. Final data and details concerning differences in pre- treatment characteristics and treatments between the two arms will be presented. SP-0430 Radiomics in radiology, what are the parameters of interest for different imaging modalities? H. Ahlström 1 1 Uppsala University, Dept of Radiology, Uppsala, Sweden CT, MRI, PET, PET-CT and PET-MRI datasets contain huge amounts of spatially detailed morphological, functional and metabolic information. Today, when analysed, these detailed datasets are typically heavily reduced to a few measurements of a priori specified measurements of interest (e.g. volumes, areas, diameters, average/maximum tracer concentrations etc.) and/or visually – and therefore inevitably subjectively – assessed by a human operator. As a result, normality/non-normality can only be assessed on these measurements and not on the entire data collected, and statistical interaction with non-imaging parameters can also be assessed only on these a priori specified measurements. In order to utilise the full potential of these image datasets, new analysis tools included in the concept Radiomics, that allow objective or quantitative assessment of all imaging data (including e.g. previously discarded information about texture), are needed. Radiomics can be divided into distinct processes: (a) image acquisition and reconstruction, (b) image segmentation and rendering, (c) feature extraction and feature qualification and (d) databases and data sharing with non-imaging data (e.g. different “omics” and clinical data) for (e) informatics analyses. Statistical knowledge of the normal range of Radiomics features are needed for the analyses. These analyses are anticipated to bring out new associations and understandings that traditional approaches could not achieve. Radiomics features can, together with non- imaging data, be included in models that have shown to Joint Symposium: ESTRO-ESR: Radiomics and imaging databases for precision radiation oncology

provide valuable diagnostic, prognostic or predictive information for oncological diseases. This information aims at improving individual patients’ outcomes by a better treatment selection. SP-0431 Radiomics in radiotherapy. How is it used to personalise treatment and to predict toxicity and/or tumour control C. Gani 1 1 University Hospital Tübingen Eberhard Karls University Tübingen, Radiation Oncology Department, Tübingen, Germany Radiomics is defined as the automated or semi-automated extraction of a large number of features from imaging datasets resulting an individual “imaging phenotype”. These features and the imaging phenotype can then be correlated with a variety of other parameters: from genetic phenotypes to oncological outcome data. Radiomics as a non-invasive procedure is of particular interest for the radiation oncologist in times of precision radiation oncology: The radiomics phenotype might help to identify patients at high risk for treatment failure and therefore candidates for more aggressive treatment. Furthermore radiomics can also be a helpful tool to predict the risk for radiation-induced toxicities and guide the dose distribution within normal tissues. This lecture will give an overview about the existing data on radiomics in the field of radiation oncology. SP-0432 Uncertainties in imaging -how they should be reported and propagated in prediction models using radiomics L. Muren 1 Aarhus University Hospital - Aarhus University, Medical Physics, Aarhus, Denmark An imaging biobank can be defined as an organised database of medical images and associated imaging biomarkers (radiology and beyond) shared among multiple researchers, and linked to other biorepositories. An imaging biobank is designed for scientific use. Image data are systematically analysed visually, manual, or (semi)- automated with the main aim to extract imaging biomarkers than can be related to patient characteristics like medical history, genomic data, and outcome or disease characteristics like genomic data, biomaterials or response to treatment. The data storage is structured in a way that the database can be queried and retrieved based on available metadata. In order to exploit the available information interactions with other databases are a perquisite. General requirements with respect to the data collection are therefore a database facilitating storage of image data and metadata, storage of derived image-based measurements, and storage of associated non-imaging data, taking into account the need to deal with longitudinal data, and to cope with multiple file formats. Finally, automated retrieval is needed for image analysis pipelines that extract image features for radiomics signatures or for hypothesis free deep learning algorithms. Abstract not received SP-0433 Imaging biobanks: challenges and opportunities A. Van der Lugt 1 1 Erasmus MC University Medical Center Rotterdam, Department of Radiology, Rotterdam, The Netherlands