ESTRO 2021 Abstract Book

S387

ESTRO 2021

Figure 1: Automatic contouring DSC in function of the volume of the organ with respect to (i) ESTRO- guidelines patients, (ii) ASTRO-guidelines patients and iii) against inter-expert variability.

Conclusion This blinded quantitative evaluation of experts’ and AI-based delineation shows that AI-driven contours when obtained after training on massive multi-centric cohorts are clinically usable. It also shows that existing quality assessment tools such as DSC fail to reflect the reality of the clinical setting and negatively bias the adoption of these methods in clinical practices. Last, but not least it also shows that inter expert variability could be more significant than the one observed between experts and the automatic solution. OC-0505 Distributed comparative patient selection in the DAHANCA35 RCT of protons vs photons for H&N cancer C.R. Hansen 1,2,3 , J. Friborg 4 , P. Skyt 3 , P. Sibolt 5 , M.S. Nielsen 6 , E.H. Samsøe 7 , A.I. Holm 8 , J. Johansen 9 , R. Zukauskaite 9 , E. Andersen 5 , M. Andersen 6 , M. Farhadi 7 , J.G. Eriksen 8,10 , J. Overgaard 10 , C. Grau 3,8 , K. Jensen 3 1 Odense University Hospital, Laboratory of Radiation Physics, Odense, Denmark; 2 University of Southern Denmark, Department of Clinical Research, Odense, Denmark; 3 Aarhus University Hospital, Danish Centre for Particle Therapy, Aarhus, Denmark; 4 Rigshospitalet, Department of Oncology, Copenhagen, Denmark; 5 Copenhagen University Hospital Herlev, Department of Oncology, Copenhagen, Denmark; 6 Aalborg University Hospital, Department of Oncology, Aalborg, Denmark; 7 Zealand University Hospital Naestved, Department of Oncology, Naestved, Denmark; 8 Aarhus University Hospital, Department of Oncology, Aarhus, Denmark; 9 Odense University Hospital, Department of Oncology, Odense, Denmark; 10 Aarhus University Hospital, Department of Experimental Clinical Oncology, Aarhus, Denmark Purpose or Objective Proton treatment for head and neck cancer has the potential to reduce dose to several important organs at risk due to the physical properties. However, there is little clinical evidence to support the theoretical gain observed in treatment plan comparisons between conventional photon IMRT and proton IMPT. The Danish Head and Neck Cancer Group (DAHANCA) decided to conduct a randomized controlled trial with an enriched patient population (clinicaltrials.gov NCT04607694). Patients with a change in normal tissue complication probability (ΔNTCP) of 5%-point or more in favour of protons for xerostomia grade 2+ and/or dysphagia grade 2+ can be randomized. A photon and a proton plan were made at the referring centre. If fulfilling the inclusion criteria, the patient was referred to the national proton centre (NPC). At the NPC, new immobilization, CT, MR scans, and a proton treatment plan were made. Identical parameters for robust optimization, setup and range uncertainties, were used. The purpose of this study was to compare the proton plan from the referring centre with the clinical proton plan at the NCP to ensure the patient selection is adequate. Materials and Methods All treatment plans for the 53 patients enrolled in the pilot phase of DAHANCA35 were collected. This included the comparative photon and proton plans generated in the local treatment centre and the clinical proton plan from the NPC. 45 patients had an oropharyngeal primary. Population-based DVHs were compared between all three plans and the stability of ΔNTCP was evaluated between the comparative and clinical proton plans. Results Of the 53 patients, 20 and 43 had a ΔNTCP gain above 5%-point for xerostomia and dysphagia respectively. Ten patients were selected for both trial arms. The median ΔNTCP for the planning comparison was 6.8% [range 5.0 to 13.3%] and 6.3% [5.3 to 10.2%] for xerostomia and dysphagia, respectively. The average change in ΔNTCP from the comparative proton plan to the clinical proton plan was -1.3% [-10.2 to 7.0%] and -1.8% [-8.3 to 5.5%] for xerostomia and dysphagia, respectively. Few patients had a substantial change in their clinical target volumes, which contributed to the change in ΔNTCP. In general, the clinical proton plans did not spare the OARs to the same level as the comparative proton plan (figure 1, boxplots). However, considerable dose gains remained between the photon and the clinical proton plans (figure 2, population-based mean DVH- plots).