ESTRO 2023 - Abstract Book

S1791

Digital Posters

ESTRO 2023

Figure1. Calculated and measured lung block thickness transmission. The error bar shows ± 2% tolerance criteria.

Conclusion The study presents the effect of simplified block thickness calculation for TBI in our department. Small lungs blocks need 0.1cm additional thickness on calculated thickness due to the scatter component from large field at extended SSD treatment.

PO-2017 Knowledge Based Planning: model iteration and plan complexity in Head and Neck patients

G. Reggiori 1,3 , A. Bresolin 2 , M. Pelizzoli 4 , S. Parabicoli 4 , A. Fogliata 4 , P. Gallo 4 , F. La Fauci 4 , N. Lambri 4,5 , F. Lobefalo 4 , P. Mancosu 4 , L. Paganini 4 , M. Scorsetti 4,4 , S. Tomatis 4 1 IRCCS Humanitas Research Hospital, Radiotherapy and Radiosurgery department, Rozzano (MI), Italy; 2 IRCCS Humanitas Research Hospital, Radiotherapy and Radiosurgery, Rozzano (MI), Italy; 3 Humanitas University, Department of Biomedical Sciences, Pieve Emanuele (MI), Italy; 4 IRCCS Humanitas Research Hospital, Radiotherapy and Radiosurgery Department, Rozzano (MI), Italy; 5 Humanitas University, Department of Biomedical Sciences, Pieve Emanuele, Italy Purpose or Objective Knowledge Based Planning (KBP) systems are becoming common tools in radiotherapy planning optimization. The KBP models are continuously updated and improved also using iterative approaches. The purpose of this work is to verify how the iterative learning process used in training KBP models affects plan quality, in terms of (i) dosimetric quality, and (ii) plan complexity and deliverability. Materials and Methods Twenty-seven Head and Neck patients treated using VMAT technique in our department were selected. A first KBP model (KBP_M1) was built using the clinically delivered plans (CP set) as training set. A second set of plans (RP1 set) from the same 27 patients was optimized using the KBP_M1. A second model (KBP_M2) was then built using the RP1 set as training set and maintaining identical geometry and set-up as the CP set. This procedure was repeated iteratively creating a third and a fourth set of plans (RP2 and RP3) and a third model (KBP_M3). CP, RP1, RP2, and RP3 sets were compared in terms of target coverage and OARs sparing. Plan complexity was evaluated with the Modulation Complexity Score (MCS). A total of 108 plans and 338 arcs were optimized and delivered. Gamma Agreement Index (GAI) (3%,2mm and 2%,1mm – 10% threshold) was calculated with Portal Dosimetry software comparing the results for the 4 sets of plans. GAI values and MCS values were reported for each plan set. Results A dosimetric improvement was observed for the RP sets compared to the CP set, reducing mean dose to OARs (i.e. 3.3, 3.2 and 2.2 Gy mean dose reduction for Oral Cavity, Larynx and Parotids, mean dose respectively) while maintaining an acceptable target coverage (i.e. V95%>95%). In the subsequent iterations the OARs mean dose reduction for the same OARs dropped to 0.8, 1.2 and 0.5 Gy for the RP2 model and 0.5, 0.9 and 0.8 Gy for the RP3 model (Fig 1) compared to the previous model. The mean MUs were increased of 5% for RP1 compared to CP and remained stable within 1% in subsequent iterations. RP sets showed a significantly lower MCS (i.e. higher complexity) compared to the CP set (p<0.05) but this complexity does not change significantly when iterating the RP models (Fig 2). GAI (3%, 2mm) was >98.8% for all plans. The analysis with stricter criteria showed a decreasing GAI moving from the CP set to the RP sets (p<0.1). The fraction of arcs with GAI(2%,1mm) <95% was 25% for the CP set and 39±3% for the RP ones. Correlation between MCS and GAI was R2=0.641.