ESTRO 38 Abstract book

S5 ESTRO 38

the future, similar frameworks are likely to be used by clinicians to dose-escalate patients at low-risk and reduce the dose for high-risk patient groups. SP-0012 Impact of AI and automation on practice L. Tagliaferri 1 1 Fondazione Policlinico Universitario Agostino Gemelli IRCCS, Radiation Oncology Department – Gemelli ART Advanced Radiation Therapy, Rome, Italy Abstract text Radiotherapy is involved in 45-55% of newly diagnosed cancer but also in the advanced stage disease thanks to the efficacy in the durable palliation and pain control. However, RT requires training and quality assurance[1]as it is a technologically complex advanced treatment[2]. To meet these needs, there will be an increased request of radiation therapy infrastructures and staffing but also the need to introduce Intelligence Artificial Guided Procedure (IAGP) and Automation in the clinical practice. The impact of this introduction will not be negligible, and it could improve overall health care quality thanks to the support of big data[3], deep learning[4], algorithmic innovation and powerful neural network models[5]. Automation using artificial intelligence is systematically involved in several fields of cancer care and, more specifically, in the radiation treatment workflow[6]. Many software aiming to improve and speed up the “contouring process” through autosegmentation[7]solutions have been released in the last years. Auto-segmentation also allows to perform the adaptive re-planning procedure during treatment[8]. Treatment planning step is also a very time-consuming task and the quality of its output is based on the physicist’s and physician’s knowledge and expertise. There is variability in this field, not only across centers, but even in the same center due to the presence of different planners. Inverse planning optimization and automated knowledge-based treatment planning approaches[9][10]improve the speed of the process and ensure that the quality of the results is optimal and not treatment planner dependent. Many AI applications have also been introduced in the radiation delivery process. Many technologies can indeed check the exact location of the target and the organs at risk improving the quality and speeding up the procedure of patient repositioning[11]and the radiation delivery[12]. Recently, Decision Support Systems (DSS) are spreading more and more: the implementation of large database andradiomics[13]provides physicians a huge amount of data that can improve clinical practice through the use of predictive models[14]and reduce the impact of knowledge gaps between domain-specific experts and non- experts[15]aiming to improve personalized cancer care. Aim of this talk is to present and discuss how the introduction of Intelligence Artificial Guided Procedure (IAGP) and Automation could impact the clinical practice and the radiotherapy workflow, especially in the fields of contouring and treatment planning, quality assurance, radiation delivery and clinical outcomes recording and prediction. It will be also analyzed how, thanks to the minimum human intervention and automatically error detection systems, automation could also provide a reduction in cure costs[16]and an escape from time- consuming repetitious tasks with significant advantages for the physicians and consequent of more time to dedicate to interaction with patients, increasing the level of humanity and patients care perception quality. Bibliography [1]Morganti AG et al. Radiother Oncol. 2008 [2]Valentini V et al. Rays. 2001 [3]Tagliaferri L et al. J Contemp Brachytherapy. 2016 [4]Damiani A et al. Eur J Intern Med. 2018 [5]Xing L et al. J. Med Phys. 2018 [6]Thompson RF et al Int J Radiat Oncol Biol Phys. 2018 [7]Naceur MB et al. Comput Methods Programs Biomed. 2018 [8]Montanaro T et al. Med Phys. 2018 [9]Good D et al. Int J Radiat Oncol

We identified a cohort of 100 patients that underwent curative hypofractionated radiotherapy to the prostate (66 Gy in 22 fractions) and retrospectively genotyped for gene dosage (copy numbers) and nucleotide polymorphisms in six different genes reported previously to be involved in response to radiotherapy: XRCC1, XRCC3, ERCC2, SOD2, VEGFA and TGFB1. Predictive models of both late rectal bleeding and erectile dysfunction, two dose limiting side effects of prostate cancer radiotherapy, were identified and cross-validated. The best performing multi-metric models contained dosimetric and biological variables, reflecting the interlinked biophysical nature of late radiation-induced toxicities. Notably, the copy number of DNA repair gene XRCC1, acting via the base excision repair pathway, and TGFB1, a key player in modulating inflammation and matrix remodelling, were found to be valuable predictive markers.

Taken together, we demonstrate that the inclusion of patient-specific mutations in DNA repair genes coupled with dosimetric parameters derived from patient-specific radiotherapy treatment plans provides a significant improvement over non-biological models. We accomplished this using an automated data-mining framework able to deal with clinical, biological, and treatment-related radiotherapy data simultaneously. In