The out-of-field neutron dose from high-energy medical LINACs is a key factor in assessing secondary cancer risk. While the neutron source strength (Q) has been characterized, data on the resulting organ-specific fast neutron equivalent dose (Hfast) for modern LINACs is scarce. This study provides new experimental data on the fast neutron equivalent dose for the Varian TrueBeam and Elekta Versa HD, including the first determination of organ-specific doses for the TrueBeam under a clinical Intensity-Modulated Radiation Therapy (IMRT) plan. We performed measurements with Solid-State Nuclear Track Detector (SSNTD, CR-39 type) in an anthropomorphic phantom using 6 MV and 18 MV beams for both standard fields and a clinical breast IMRT plan. We calculated the equivalent dose using two distinct methodologies: Linear Energy Transfer (LET) spectrometry and a direct calibration factor approach. Our results quantify the significant dose variations between organs, driven by tissue composition, and confirm a measurable, non-trivial dose from 6 MV beams, which is relevant for high-dose stereotactic techniques. These novel experimental data provide essential benchmarks for validating the models used in clinical risk assessment.

Accurate characterization of the neutron source strength (Q) is essential for radiation protection and risk assessment in high-energy radiotherapy. This study provides a comprehensive experimental and Monte Carlo characterization of Q values for three widely used linear accelerators: the Varian Clinac 2100C, Varian TrueBeam, and Elekta Versa HD. We performed experimental measurements using CR-39 detectors under clinically relevant conditions and developed a validated GATE Monte Carlo model for the Clinac 2100C to compare methodologies. This work presents the first reported experimental Q values for the TrueBeam at both 6 MV and 18 MV, and for the Versa HD at 18 MV. Our results reveal significant model-dependent variations, with the Varian LINACs producing higher Q values than the Elekta Versa HD. We also demonstrate that the field collimation method (jaws vs. MLCs) is a critical factor, with MLC-defined fields capable of increasing neutron production significantly. Furthermore, we confirmed measurable, non-zero Q values at 6 MV, a finding relevant for high-dose stereotactic treatments. These model-specific benchmark data are essential for improving the accuracy of clinical risk assessments and shielding calculations in modern radiotherapy facilities.

Over the last decades, the use of artificial intelligence, machine learning and deep learning in medical fields has skyrocketed. Well known for their results in segmentation, motion management and posttreatment outcome tasks, investigations of machine learning and deep learning models as fast dose calculation or quality assurance tools have been present since 2000. The main motivation for this increasing research and interest in artificial intelligence, machine learning and deep learning is the enhancement of treatment workflows, specifically dosimetry and quality assurance accuracy and time points, which remain important time-consuming aspects of clinical patient management. Since 2014, the evolution of models and architectures for dose calculation has been related to innovations and interest in the theory of information research with pronounced improvements in architecture design. The use of knowledge-based approaches to patient-specific methods has also considerably improved the accuracy of dose predictions. This paper covers the state of all known deep learning architectures and models applied to external radiotherapy with a description of each architecture, followed by a discussion on the performance and future of deep learning predictive models in external radiotherapy.

Introduction. In the past few years, the development of complex radiotherapy techniques has led to the development of new dosimetric detectors based on gel, such as Fricke or polymer gels, to enhance quality control. They are able to provide 3D absorbed dose distributions with a good spatial resolution for photon radiotherapy treatment, but additional studies are needed for protontherapy. The gel studied in this work must take into account the irradiation conditions and the reading mode. The great sensibility of VIPAR gel in high dose range [ 1 Kozicki et al. On the development of a VIPARnd radiotherapy 3D polymer gel dosimeter. J Phys Med Bio. 2017; 60: 986 Crossref Scopus (44) Google Scholar , 2 Kozicki et al. On the development of the VIPAR polymer gel dosimeter for three-dimensional dose measurements. J Macromol Symp. 2007; 254: 345-352 Crossref Scopus (33) Google Scholar ] makes VIPAR gel a good candidate for proton irraditions (sensitive until 120 Gy).

Introduction During these last years, treatment planning system (TPS) algorithms that compute the absorbed dose have improved. A high precision between delivered absorbed dose and prescribed dose is required, limiting the uncertainties up to 5% between delivered dose and prescribed dose according to the ICRU recommendations [ [1] ICRU. Determination of absorbed dose in a patient irradiated by beams of x or gamma rays in radiotherapy procedures, Report 24. J Int Commission Radiat Units Measure os13, (Oxford University Press, 1976). Google Scholar ]. One of the parameters contributing to these uncertainties is the couch top, which attenuates the beam from 2% to 15% when it passes through [ [2] Olch A.J. Gerig L. Li H. Mihaylov I. Morgan A. Dosimetric effects caused by couch tops and immobilization devices: report of AAPM Task Group 176. Med Phys. 2014; 41: 1-30 Google Scholar ]. Last generation's carbon fiber couch tops act as a bolus material, increasing skin dose while decreasing depth-dose. Therefore, the consideration of the couch top by the TPS is important for avoiding bad tumor control, overdose of organs at risk and skin radiotoxicity effects [ [2] Olch A.J. Gerig L. Li H. Mihaylov I. Morgan A. Dosimetric effects caused by couch tops and immobilization devices: report of AAPM Task Group 176. Med Phys. 2014; 41: 1-30 Google Scholar ]. Methods The reference technique for the dose distribution calculations is based on Monte Carlo method (MC) [ [3] Mohan R. Antolak J. Hendee W.R. Monte Carlo techniques should replace analytical methods for estimating dose distributions in radiotherapy treatment planning. Med Phys. 2001; 28: 123-126 Crossref PubMed Scopus (21) Google Scholar ]. Among the most well-known codes, BEAMnrc/EGSnrc allows to model the accelerator treatment head and to use CT images (DICOM format) in order to simulate the dose distribution to voxelized phantoms. Furthermore, the CT data have to be converted to densities and to the corresponding materials, taking into account the couch top [ [4] Verhaegen F. Devic S. Sensitivity study for CT image use in Monte Carlo treatment planning. Phys Med Biol. 2005; 50: 937-946 Crossref PubMed Scopus (82) Google Scholar ]. A software developed by the team using a conversion ramp between Hounsfield Units (HU) and densities/materials, specific to the couch top, will be used to create the voxelized phantoms. Results MC method takes into consideration not only voxel densities, but also materials, if a wrong material assignment is made, it could lead to dosimetric errors in the dose distribution calculations. Different treatment sites will be considered in Volumetric Modulated Arc Therapy (VMAT). Phantoms with the couch top will be created and dose distributions will be compared to TPS dose distributions. Conclusions The results of this study contributed to evaluate different methods of taking into account the couch top, using MC.

In radiation therapy, the trade-off between accuracy and speed is the key of the algorithms used in Treatment Planning Systems (TPS). For photon beams, commercial solutions generally relies on analytic algorithms, biased Monte Carlo, or heavily parallelized Monte Carlo on Graphics Processing Units (GPU). Alternatively, we propose an algorithm using Artificial Neural Network (ANN) to compute the dose distributions resulting from ionizing radiations inside a phantom [1] , [2] . We present an evolution of this platform taking into account modulated field sizes and shapes, and various orientations of the beam to the phantom. Firstly, tomodensitometry-based phantoms are created to validate the dose distribution computed for a square beam in heterogeneous areas (head and neck, lungs). Secondly, IMRT treatments are simulated in these phantoms. To validate our approach, we compare our results with the Analytical Anisotropic Algorithm (AAA) and Monte Carlo simulations. Cross-comparisons are performed for square beams and IMRT treatments. The dose distributions are evaluated using gamma indices and profile extractions. The dose distributions computed from IMRT treatments require less than two minutes using a standard Central Processing Unit (CPU). We aim at providing a fast and accurate solution for TPS quality control.

Introduction: Monte Carlo simulations are the most accurate methods for dose calculations in media with complex geometry. Nowadays, despite the development of fast computational systems, Monte Carlo methods without bias are still not usable for clinical routine. Hence, Monte Carlo simulations are restricted to research. The following study presents the modeling of the TrueBeam Novalis STX linear accelerator (LINAC) of Varian using BEAMnrc/DOSXYZnrc Monte Carlo code. Such work has already been performed by different researcher teamsusing either BEAMnrc/DOSXYZnrc or other available Monte Carlo codes such as PENELOPE, Geant4/GATE and MCNPX [ 1 Bergman A.M. et al. Monte Carlo modeling of HD120 multileaf collimato on Varian TrueBeam linear accelerator for veriffication of 6X and 6X FFF VMAT SABR treatment plans. J Appl Clin Med Phys. 2014; 15: 4686 PubMed Google Scholar , 2 Gete E. et al. A Monte Carlo approach to validation of FFF VMAT treatment plans for TrueBeam linac. Med Phys. 2013; 40: 021707 Crossref Scopus (32) Google Scholar ]. Nevertheless, the specificities of each single machine combined with a very little information exchange between different medical physic teams implies that every group needs to perform its own LINAC modeling.

Introduction In order to safely administrated highly complexes radiotherapy technics like VMAT or IMRT, the implementation of a quality assurance software is in progress at Chrono-environnement UMR CNRS 6249 laboratory based in Montbeliard (France). The Monte Carlo modeling of the TrueBeam Novalis STx used in Montbeliard hospital was performed. The simulation of particles transport through jaws, MLC and water phantom was performed using the Varian generic phase-space profiles for the 10 MV beams in both flattened and unflattened (flattening filter free) mode. Hereafter, the validation of the modeling for clinical quality assurance program is presented. Methods The modeling of TrueBeam Novalis STX of Varian was implemented using the Monte Carlo code BEAMnrc/DOSXYZnrc. BEAMnrc was used to create field specific phase spaces under the jaws for field ranging from 3 cm * 3 cm to 20 cm * 20 cm. DOSXYZnrc was used to calculate doses in water phantom. Calculated cross-lines profiles at various depths, percent depth doses and output factors were compared with gold standard measurements provided by Varian. The simulation and measurement were compared using the gamma-index analyses method. Results For cross-line profiles, agreements in the penumbra region (80–20% width) and in the field size defined at 50% were better than 2 mm for all field sizes. Moreover agreement better than 1.5%, 1 mm and 2%, 2 mm were found for 10 MV-FFF and 10MV, respectively. Calculated and measured percent depth doses beyond the buildup region agreed within 1.5%, 1 mm for 10 MV-FFF beams and within 2%, 2 mm for 10 MV beams. The agreement between calculate and measured output factors for all fields were within 1%. Furthermore the 10 MV-FFF results show less noise compare to the 10 MV’s. Conclusions The X10 and X10-FFF Varian phase-spaces profiles have been validated in homogeneous medium. The perspectives of this study include the validation of the modeling in a heterogeneous medium. And afterward, the validation of a treatment plan through a comparison between a Monte Carlo VMAT simulating plan and a TPS one.

Introduction : In order to safely administrated highly complexes radiotherapy technics like VMAT or IMRT, the implementation of a quality assurance software is in progress at Chrono-environnement UMR CNRS 6249 laboratory based in Montbéliard (France). The Monte Carlo modeling of the TrueBeam Novalis STx used in Montbéliard hospital was performed. The simulation of particles transport through jaws, MLC and water phantom was performed using the Varian generic phase-space files for the 10MV beams in both fattened and unflattened (flattening filter free) mode. Hereafter, the validation of the modeling for clinical quality assurance program is presented.

Introduction In radiation therapy, one of the most time consuming step is due to the Treatment Planning System (TPS) and quality control algorithms for the dose computations. Using Artificial Neural Network (ANN), it is possible to compute the doses inside a phantom resulting from ionizing radiations [ 1 Mathieu R. Martin E. Makovicka L. Gschwind R. Contassot-Vivier S. Bahi J. Calculations of dose distributions using a neural network model. Phys Med Biol. 2005; 50: 1019-1028 Crossref PubMed Scopus (15) Google Scholar , 2 Vasseur A. Makovicka L. Martin E. Sauget M. Contassot-Vivier S. et al. Dose calculations using artificial neural networks: a feasibility study for photon beams. NIM. B. 2008; 266: 1085-1093 Crossref Scopus (17) Google Scholar ]. Material and methods More precisely, once the ANN is trained, it can compute a dose for every voxels of the phantom according to the beam characteristics. The main advantage is the computation time: the time-consuming step is the training of the ANN, that has to be performed only once and offline. The execution time, i.e. the dose associated to a voxel, is very low. We present an evolution of the platform presented in [ [3] Sauget M. Laurent R. Henriet J. Salomon M. Gschwind R. et al. Artificial neural networks–ICANN 2010. Springer, Berlin Heidelberg2010: 261-266 Google Scholar ], to take into account various field sizes and shapes, and also consider different orientation of the beam to the phantom. To validate the ANN computations, we compare them with Monte Carlo simulations. Homogeneous phantoms are used to validate interpolations on various densities and depth. Then, realistic phantoms are created to validate the dose distribution computed for a square beam in complex areas (head and neck, lungs, prostatic area). Finally, IMRT treatment simulations are validated. Results We compare the results obtained using our approach, the Eclipse TPS (AAA), and Monte Carlo simulations for a 10 cm x 10 cm photon beam on phantoms extracted from clinical data. The same procedure is applied to validate IMRT treatments. To compare the dose distributions, 3D gamma indices are computed, as well as profile extractions. For IMRT treatments, the dose distribution is computed in less than one minute on a traditional computer. Conclusions We have demonstrated the performance of our dose distribution engine for IMRT. Our goal is to provide a fast and accurate solution for TPS quality control. Therefore, we are currently working on the development of arctherapy treatment simulations.

Introduction: In radiation therapy, various quality controls are involved to ensure that the dose delivered to the patient during treatment corresponds to the prescribed and planned dose. Thus, in intensity modulated radiation therapy (IMRT), the Electonic Portal Imaging Device (EPID) is used to verify treatment plans calculated by treatment planning systems (TPS). During these controls, some differences can be observed between the measurement made by the EPID and the computation made by the TPS. To understand these differences, a theoretical study using Monte Carlo simulation was performed.

Introduction Currently, S factor computations in internal radiation therapy or TPS quality controls in external radiation therapy are only performed in 3D, i.e. without taking into account physiological or metabolic movements. That is why we propose to add a dimension representing time to the IRSN phantoms and thus enable these protocols in 4D. Materials and methods The NEMOSIS platform (based on an Artificial Neural Network – ANN) has already been presented and validated in previous works, which were dedicated to the customized simulation of internal lung motions. To adapt it to phantoms that were determined according to anthroporadiametric data, new entries (perimeter and height of a cylinder representing the lungs) were added to the ANN. The IRSN phantoms (12 phantoms of heights varying between 165 and 185 cm) only account for the organ contours. To make NEMOSIS able to proceed these contours, we have elaborated an algorithm that automatically tracks the evolutions of the lung contours of each patient on every 4DCT. After the learning step of the patient data, our platform is used to simulate the motion of phantoms. Results Thanks to this approach, 32,480 points were computed and added to our dataset, which is constituted of 16 patients over 10 respiratory phases. The similarity index, calculated between the simulated volumes and 4D, is strictly greater than 0.94 and thus allow the validation of our approach applied to a test patient. The motion of the phantoms computed by NEMOSIS is consistent and represents realistic variations according to the lung localization and phase. The hysteresis is also depicted, but measured variation of the lung volume for this phantom is of only 0.185 l. Conclusion Perspectives: Despite very promising results, the quality of the motion simulation can still be improved by taking into account the deformation of the diaphragm. However, our 4D simulation of the lung motion of phantoms can already be validated.

Mastering the sorting of the data in signal (nD) can lead to multiple applications like new compression, transmission, watermarking, encryption methods and even new processing methods for image. Some authors in the past decades have proposed to use these approaches for image compression, indexing, median filtering, mathematical morphology, encryption. A mathematical rigorous way for doing such a study has been introduced by Andrei Nikolaievitch Kolmogorov (1903-1987) in 1957 and recent results have provided constructive ways and practical algorithms for implementing the Kolmogorov theorem. We propose in this paper to present those algorithms and some preliminary results obtained by our team by applying them to image processing problems such as compression, progressive transmission and watermarking.

The authors present a novel approach for image compression based on an unconventional representation of images. The proposed approach is different from most of the existing techniques in the literature because the compression is not directly performed on the image pixels, but is rather applied to an equivalent monovariate representation of the wavelet-transformed image. More precisely, the authors have considered an adaptation of Kolmogorov superposition theorem proposed by Igelnik and known as the Kolmogorov spline network (KSN), in which the image is approximated by sums and compositions of specific monovariate functions. Using this representation, the authors trade the local connectivity and the traditional line-per-line scanning, in exchange of a more adaptable and univariate representation of images, which allows to tackle the compression tasks in a fundamentally different representation. The contributions lie in the several strategies presented to adapt the KSN algorithm, including the monovariate construction, various simplification strategies, the proposal of a more suitable representation of the original image using wavelets and the integration of this scheme as an additional layer in the JPEG 2000 compression engine, illustrated for numerous images at different bit rates.