ESTRO 2020 Abstract book

S765 ESTRO 2020

A machine learning algorithm was developed and tested to validate the concept of predicting electron beam parameters from profile data. This method would allow significant time gain in the context of integrating multiple accelerator models in a Monte Carlo-based dose verification system. PO‐1353 Validation of a commercial software for in vivo patient Quality Assurance C. Talamonti 1,2 , L. Marrazzo 2 , S. Russo 3 , E. Vanzi 4 , C. Arilli 2 , M. Casati 2 , S. Pini 3 , A. Ghirelli 3 , P. Bastiani 5 , G. Simontacchi 6 , S. Pallotta 1,2 , M. Esposito 3 1 University of Florence, Dip Scienze Biomediche Sperimantali e Cliniche "Mario Serio", Firenze, Italy ; 2 Azienda Ospedaliero Universitaria Careggi, SOD Medical Physics, Florence, Italy ; 3 Azienda Sanitaria USL Toscana Centro, S.C. Fisica Sanitaria Firenze, Florence, Italy ; 4 Azienda Ospedaliero Universitaria Senese, S.C. Fisica Sanitaria Siena, Florence, Italy ; 5 Azienda Sanitaria USL Toscana Centro, S.C. Radioterapia Firenze, Florence, Italy ; 6 Azienda Ospedaliero Universitaria Careggi, SOD Radioterapia, Florence, Italy a patient quality assurance software for in-vivo dose verification using the EPID measured fluence of the treatment fields. Back- projection models are used to reconstruct the 3D dose distribution in the CT planning model of the patient. The aim of this study was to test the accuracy of the two dose calculation algorithms available in DC: the pencil beam (PB) and the collapsed cone convolution (CCC). Material and Methods The accuracy of the algorithms was tested on phantom in the steps of entrance fluence estimation and dose calculation. Measured OFs and PDDs were evaluated respect to the ones reconstructed by DC using both algorithms. These tests were performed on IBA “I’m RT” phantom. Its central cubic insert was filled in three different ways: homogeneously with RW3 slabs (phantom H) or replacing the central slabs with air (phantom A) or bone inserts (phantom B). An anthropomorphic phantom study was also performed. Four VMAT clinical plans (head-neck, CNS, prostate and lung) were calculated using the TPS Monaco 5.11 on the RANDO phantom and delivered by an Elekta Synergy. During the delivery a Gafchromic® EBT3 film was embedded in the Rando at the isocenter plane. Each plan was delivered three times. Films were registered with TPS and DC dose maps. Gamma analysis (3%, 2mm, global) and isocenter dose were used to compare measured dose distribution with respect to DC calculated ones using both algorithms. Results In-vivo measurements produced relative differences for the phantom H and B, on OFs ranging between -0.48 and 0.92% (0.34%), and between 0.70 and 0.80% (0.30%) respectively. Much higher differences were found in phantom A owing to the use of PB, especially for small fields where a 64.2% difference was found for the 2.4 cm field size. The differences decrease by increasing the field size: a 2.1% difference was found for 15.2 cm field size. The average difference for phantom A was 16.23%. For the PDDs the percent difference for H and B phantoms were respectively 1% and 1.5% at all depths excluding build up region. For phantom A, DC with PB overestimates the dose up to 12% and 110% in the air slab for the 10.4 cm and 2.4 cm field size respectively. In fig. 1a are shown the PDDs calculated with DC PB and CCC and Monaco TPS relative to the field 10.4x10.4cm 2 . Results on anthropomorphic phantom are reported on fig1b. Except for the lung, in the other sites there is no difference in the use of the two algorithms. A large sigma value in the brain plan is due to an air cavity near the target. These values are in agreement with the previous Purpose or Objective DosimetryCheck (DC) is

series of simulated dose profiles and percentage depth doses (PDD) are created from the MC model for the 115x100mm MLC field, while varying electron beam spot size (from 1 to 4 mm) and energy (from 4 to 8 MeV). A Ridge regression algorithm is first trained to predict energy and spot size by splitting the simulated curves into training and test sets, allowing to finetune the parameters of the algorithm. The ML model is then applied on measured data provided by five other M6-equipped institutions, and the predicted values are introduced in a MC model to generate simulated profiles for the 115x100mm MLC and 5 mm fixed beams, to be compared with the measurements.

Figure 1. Principle of the electron beam characteristics prediction. Results For our M6 device, optimal agreement between simulated and measured profiles in the water tank was reached for a monoenergetic electron beam of 6.75 MeV with a gaussian spatial distribution of 2.4 mm FWHM. Re-calculation of patient plans showed a good agreement (<2%) between the TPS algorithms and Moderato. During the optimization of the ML algorithm, cross-validation showed that electron beam energy and spot size could be predicted with a mean absolute error (MAE) of 0.1 MeV and 0.3 mm respectively. Predicted values for the five other M6 systems varied significantly from one device to another (Table 1). Simulations generated from these predicted values demonstrated a close agreement (<3%) between simulated and measured dose curves, except for centre #1 where differences went up to 6% (figure 2), which might be due to an asymmetry in the measured dose profiles for that specific machine. Table 1. Predicted electron beam spot sizes and energies for the five M6 devices. Centre Predicted spot size (mm) Predicted energy (MeV) #1 2.7 7.1 #2 1.9 6.8 #3 1.8 6.6 #4 3.4 6.5 #5 2.0 6.5

Figure 2. Dose profiles for the five M6 devices, measured (red lines) and simulated (blue dots) for the 115x100mm MLC (top) and the fixed 5mm (bottom) beams. Conclusion