ESTRO 2024 - Abstract Book

S3449

Physics - Dose calculation algorithms

ESTRO 2024

Monica Bianchi 1,2 , Nicola Lambri 2,3 , Andrea Bresolin 2 , Simone Buzzi 1,2 , Damiano Dei 2,3 , Pasqualina Gallo 2 , Francesco La Fauci 2 , Francesca Lobefalo 2 , Lucia Paganini 2 , Sara Parabicoli 1,2 , Marco Pelizzoli 1,2 , Giacomo Reggiori 2 , Stefano Tomatis 2 , Caterina Zaccone 1,2 , Marta Scorsetti 2,3 , Cristina Lenardi 1,4 , Pietro Mancosu 2 1 Università degli Studi di Milano, Physics, Milan, Italy. 2 IRCCS Humanitas Research Hospital, Radiotherapy and Radiosurgery, Rozzano, Milan, Italy. 3 Humanitas University, Biomedical Sciences, Milan, Italy. 4 National Institute for Nuclear Physics, Milano Division, Milan, Italy

Purpose/Objective:

Modern radiotherapy (RT) planning often involves the co-registration of CT and MRI scans to benefit from the complementary advantages of the two imaging techniques. To streamline this process, MRI-based RT has recently been proposed, and the creation of synthetic CTs from MRI images through the use of AI algorithms is showing promising results. Although CT data is stored in a 12-bit format, utilizing 8-bit CTs to train these algorithms can be more computationally efficient. However, the resampling of the original scans introduces uncertainties in dose calculation. The aim of this study is to investigate the dosimetric and clinical impact of a non-linear look-up-table (LUT) for resampling whole-body CTs from 12 bits to 8 bits in total marrow irradiation (TMI) plans.

Material/Methods:

Ten patients treated with TMI were randomly selected from the internal database. The original CTs (oCTs) consisted of acquisitions from the patient’s head to mid femurs, and were resampled using a linear and a non-linear LUT. The latter was implemented as a piecewise function to enhance the HU values of clinical interest, achieving a discretization of 2HU within the range [0,100] HU, with a non-linearly increase in discretization up to 100HU for densities >1650HU. Six treatment plans per patient were created on the Eclipse TPS (Varian), employing the analytical anisotropic algorithm (AAA) and AcurosXB (AXB) on oCT, non-linearly resampled CTs (rCT_NL) and linearly resampled CTs (rCT_L), with dose distributions recalculated using identical parameters of the clinical plan. The resampling effect on the fidelity of the HU values was quantified by computing the root mean squared error (RMSE) between oCTs and resampled CTs (rCTs). The dosimetric evaluation was performed using a paired t-test (p<0.05 significant) by comparing the D98% and D2% between oCT-rCT recalculated with the same algorithm, and between oCT-oCT recalculated with AAA and AXB. PTV, bony structures inside it, brain, liver, and lungs were considered for this analysis. The dose distributions global agreement was evaluated by computing the 3D gamma passing rate (GPR) with 1%/1.25 mm acceptance criteria and 20% threshold using pymedphys [1]. Finally, 6 radiation oncologists rated on a 1-5 scale the image reliability of rCTs for contouring and diagnostic purposes on various organs and districts. The Wilcoxon signed rank test was used to assess whether oCTs and rCTs were distinguishable (p<0.05 significant).

Results:

The HU comparison between oCT-rCT revealed a higher average RMSE when using a non-linear LUT as opposed to a linear one, with values of 10±1 HU and 7.5±0.6 HU, respectively. However, the non-linear LUT ensured better dosimetric agreement for both algorithms (Figure 1). Overall, the resampling increased the hotspots in the dose distributions, with the non-linear resampling producing smaller differences (around 0.1% for D98% and D2%) than the linear one (0.2%), except for the lungs (respectively, up to 0.5% and 0.3%). For PTV, soft tissues and lungs, the dose variations arising from the use of rCTs were less relevant than those coming from a change in the calculation algorithm (up to 1.8% for D98% and D2%), while this was not the case for bones (around 0.2%). Despite maximum