ESTRO 2021 Abstract Book

S1449

ESTRO 2021

Figure 1: (a) MAE/Dose per Fraction (b) comparison of planar doses measured in the detector (red circled) and predicted by the VGG deep neural network model (green circled)

Conclusion In this study we present a novel Deep Learning methodology for 3D voxel-by-voxel dose prediction, able to reproduce results in HT. The method is promising and open the possibility for improving over current state-of- the-art DVH-based prediction and planning techniques. References [1] G. Valdes et al., Med Phys 43(7) 2016 [2] S Yartsev, T. Kron, and J .Van Dyk, Biomed Imaging Interv J. 2007 Jan-Mar; 3(1): [3] M. J. Nyflot, P.s Thammasorn, W. Art Chaovalitwongse, Med. Phys. 46 (2) 2018. Acknowledgements Authors would thanks Mark Geurts and Luca de Carli for their help and support on machine specific information. Corresponding Author: carlotti.dan@gmail.com PO-1724 Reducing the number of incidents that occur during patient treatment K. Pasma 1 , R. Meerveld 2 , J. Barnhoorn 3 , M. Leusink 3 1 Radiotherapiegroep, Medical Physics, Arnhem, The Netherlands; 2 Radiotherapiegroep, Radiation Oncology, Arnhem, The Netherlands; 3 Cablon Medical, Development, Leusden, The Netherlands Purpose or Objective Radiation therapy is a highly regulated medical practice with historically low error and injury rates. However, when errors do occur, they can result in suboptimal treatment, especially when misadministration results in a dose unintentionally delivered to vital organs or structures, such as the spinal cord, heart, lungs, or brain. Delivering radiation therapy is a team effort requiring collaboration and clear communication between the radiation oncologist, medical physicist, dosimetrist, and radiation therapist/technologist. Preventing errors in the delivery of radiation therapy involves not only understanding and appropriately utilizing new advances in technology, but also utilizing established patient safety procedures that optimize safe healthcare delivery. In this study we demonstrate how we effectively reduced the number of incidents in the clinic by analyzing the incidents that did occur and implementing hard- and software to detect conditions just prior to patient treatment that could lead to mistreatment. When detected the system blocks the treatment and warns the RTT. Materials and Methods Since 2007, reported incidents as well as incidents that did not lead to a treatment error have been analyzed. Adapting the idea of the Swiss cheese model [Reason, Phil Trans R Soc B 327:475–84,1990], we identified the geometrical risks during radiation therapy. If possible, a risk is mitigated with an interlock implemented in CNERGY Check (Cablon Medical, Leusden, The Netherlands). This system is integrated with our RV system (Mosaiq 2.83), CBCT system (XVI), existing interlock system, patient verification system and readout of the couch pedestal. It prevents delivering a radiation dose or acquiring a CBCT if an erroneous condition in the setup is detected at the same time warning the RTT. All ten linear accelerators (Elekta, Crawly, UK) from our hospital are equipped with the CNERGY system. Results In total, fifteen risks were identified, for example (i) beam delivery started prior to completing the IGRT procedure, (ii) delivering a treatment fraction for patient A to patient B (2 in 10 year), (iii) preparing the patient for treatment while the plan has not been sent to the CBCT system, (iv) verifying if the isocenter in the R&V system is equal to that in the CBCT system, (v) incorrect couch position after repositioning the patient between two treatment beams. For these situations CNERGY Check will trigger an interlock preventing from proceeding with a patient workflow under erroneous conditions until the RTT has fixed the cause. This has resulted in significant reduction of the number and severity of incidents. Some identified risks have been completely eliminated. Based on new incident reports we strive to constantly improve detection and prevention of erroneous conditions.