ESTRO 2022 - Abstract Book

S25

Abstract book

ESTRO 2022

irradiated gel tube to predict the slope, or sensitivity of the calibration curve. These features were extracted using Pyradiomics package. The selection of most important features for the target and the forest hyperparameter tuning was also made. A division of 70% of the data for train the model and 30% for test was applied in this work. An overview of the proposed methodology is shown on Figure1. Figure 1- Work methodology

Results 14 features were selected as most important for the target prediction. After the selection of the best forest parameters, the model training, and its application to the test data, a plot of the predicted slope, versus the real (true) value was done (figure 2). The model´s performance was evaluated with mean square error (MSE) and the data uncertainty deviation (Table 1).

Figure 2- Graphic involving the predations and the uncertainty deviation Table 1- Model evaluation MSE uncertainty deviation

uncertainty deviation

3,7 x 10 -4

± 0.03

± 0.04

80% of the data

100% of the data

Conclusion The regression model proved to be satisfactory for the prediction of the calibration curve slope in gel dosimetry, with low MSE and uncertainty deviation of ± 0.04 including 100% of de data.

MO-0052 The impact of ion chamber components on kB,Q for reference dosimetry in MRgRT

J. Navarro Campos 1 , J. de Pooter 2 , L. de Prez 2

1 LUMC, Radiotherapy, Leiden, The Netherlands; 2 VSL Dutch Metrology Institute, Ionising radiation, Delft, The Netherlands

Purpose or Objective For reference dosimetry in MRgRT, k B,Q is used to correct for the impact of the magnetic field on the chamber calibration coefficient. It has been demonstrated that for accurate simulation of k B,Q the dead volume (DV) must be considered. This work goes one step further by analysing the different contribution of all chamber components to k B,Q .

Materials and Methods