ESTRO 2021 Abstract Book

S1423

ESTRO 2021

Physics and Applied Mathematics, Pamplona, Spain

Purpose or Objective To describe our experience to commission a proton therapy system and to present its validation using several

fields and clinical cases. Materials and Methods

Our proton therapy system is model ProBeat-CR, manufacturated by Hitachi. The synchrotron extracts beams between 70.2 and 228.7 MeV. The TPS is RayStation (RaySearch Laboratories AB). The validation of the model payed special attention to the Monte Carlo correction made by the TPS to the measured IDD for the lack of signal in the Bragg peak chamber, the adequacy of the spot modeling with one circular Gaussian, and the CT calibration. We measured for all 98 energies the IDD using a Bragg Peak chamber PTW 34070 assembled in a PTW MP3-L water tank, spot profiles at several distances with an IBA Lynx scintillator, and absolute dose values with an Advanced Markus (PTW 34045) plane parallel ionization chamber. All ionization measurements of IDD were converted to dose by using the stopping power ratio fit provided in TRS-398. For validation, we measured square fields involving several energies (SOBP), with sizes ranging between 2 x 2 cm 2 and 30 x 30 cm 2 at different depths (between 5 and 20 cm) and SOBP widths (5 and 10 cm). Absolute dose values were gotten in the middle of SOBP, as well as along the depth (in central-axis PDD locations). We used two ionization chambers (Farmer and Advanced Markus). Calculation and measurement of the spot size and shape in water using a synthetic diamond (PTW 60019) was also performed. We used radiochromic film and IBA 2D MatrixXX detector to measure the dose of seven clinical fields in acrylic. Inhomogeneities were handled through the CT calibration. We did a stoichiometric CT calibration and tested it on a variety of biological Agreement in the absolute dose in the middle of SOBP (Farmer measurement) is around 1%, but increases to 2% for the 3 x 3 cm 2 field and to 3% for the 2 x 2 cm 2 . In the measurements along depth (PDD, Advanced Markus measurement), average differences are around 0.7%. Comparisons of absolute dose in similar situations involving range shifter was slightly worse (0.5%) for those similar plans except including the range shifter. All gamma index comparisons for a series of clinical plans for different treatment sites (3%, 3mm) is mostly above 95% and always above 90%. SPR predicted using our calibration showed an excellent agreement (better than ±1%) with measured values. Conclusion RayStation Monte Carlo modelling for proton beams generated with our synchrotron predicts dose with great accuracy. SPR derived values matched those measured experimentally. Patient plans are accurately calculated. tissues. Results PO-1697 Gradient-based neuroevolution of augmenting topologies for compact, low compute deep ANN search S. Thulasi Seetha 1,2 , K. Driessens 3 , H. Woodruff 1 , T. Rancati 4 , E. Bertocchi 4 , U. Pastorino 2 , P. Lambin 1 1 GROW - School for Oncology and Developmental Biology, Maastricht University Medical Centre, Department of Precision Medicine, Maastricht, The Netherlands; 2 IRCCS Foundation National Cancer Institute, Department of Thoracic Oncology, Milan, Italy; 3 Maastricht University, Data Science & Knowledge Engineering, Maastricht, The Netherlands; 4 IRCCS Foundation National Cancer Institute, Prostate Cancer Unit, Milan, Italy Purpose or Objective To propose a novel fast, compact, and low compute hybrid neural architectural search (NAS) algorithm which we call gNEAT for complex deep learning applications; validate the algorithm, and apply it for the task of lung gNEAT essentially searches over a space of possible network solutions starting from the most basic architecture (containing only the input and output nodes). During the course of the search, additional hidden nodes and connections are introduced incrementally. Compactness in the context of ANNs is measured based on the minimum number of parameters (nodes and connections) needed to solve a task, in relation to some known optimal solution. Validation of gNEAT's compactness is done by XOR and Autoencoder (AE) tasks (see Fig 1.a). Another important quality of gNEAT is that it can perform complex NAS with incredible speed and low computational cost. This is achieved by reducing the population size while ensuring diversity within the population for sufficient search space exploration. Furthermore, to offset the complexity associated with gradient-descent training, computation saving techniques such as the use of a proxy dataset (using resized dataset) and a blueprint (using a predefined structure and searching its components) are integrated (see Fig 1.c). Once the components are evolved, the resultant model is then applied to the original dataset. gNEAT’s abilities for complex NAS are showcased by evolving a 2D UNet architecture for lung cancer segmentation (see Fig 1.b) on the Medical Segmentation Decathlon (Lung D6) dataset. The results are then compared to the state-of-the-art (STOA) nnUNet pipeline (re-implemented). Even though nnUNet 3D holds the current STOA benchmark on this dataset, we will be limiting all the experiments to nnUNet 2D architecture to reduce complexity. In addition to this, we perform an external validation of nnUNet 2D and the evolved solution on the open-source NSCLC Radiomics (Lung 1) dataset. cancer segmentation. Materials and Methods