ESTRO 2022 - Abstract Book

S1468

Abstract book

ESTRO 2022

PO-1669 Robustness of neural networks for corrections of low-resolution and noisy small photon beam profiles

A. Schönfeld 1 , G. Yan 2 , B. Poppe 3 , H.K. Looe 3

1 Physikalisch-Technische Bundesanstalt PTB, 6.2 Dosimetry for Radiation Therapy and Diagnostic Radiology, Braunschweig, Germany; 2 University of Florida, Department of Radiation Oncology, Gainesville, USA; 3 University of Oldenburg, University Clinic for Medical Radiation Physics, Medical Campus Pius Hospital, Oldenburg, Germany Purpose or Objective Artificial neural networks (NN) are used to deconvolve transverse dose profiles of small radiation fields perturbed by the detector volume effect. This study investigates the robustness of a three-layer NN architecture with regards to varying sampling distances and signal-to-noise ratios (SNR) of the input data. Materials and Methods Transverse profiles were acquired using a Semiflex 3D 31021 and, as reference, a microDiamond detector 60019 (both PTW- Freiburg) for dosimetric field sizes ranging from 0.56 x 0.56 cm ² to 4.03 x 4.03 cm ² and a constant sampling distance of 0.3 mm. The high resolution and high SNR data have been used to train a reference NN model. Firstly, based on the Nyquist theorem, the maximum required sampling distance for each investigated field size has been derived. Secondly, the Semiflex 3D measurements were down-sampled to different sampling rates higher, equal to, or lower than the Nyquist frequency. In combination with two interpolation methods (linear and spline), individual NN models were trained on data with varying resolution to deconvolve and up-sample the measurements matching that of the high resolution reference profiles. Lastly, the robustness of the reference NN model against measurement noise was studied by adding artificial white Gaussian noise to the Semiflex 3D measurements. Subsequently, NN models were trained with the low SNR data. The results obtained using these retrained NN models were compared to the reference NN model. Results The maximum required sampling rate according to the Nyquist theorem was found to increase with the field size. For the case where linear interpolation was used, the retrained NN models show poorer performance even though the profile was sampled at a sampling rate higher than the Nyquist frequency. In contrast, NN models retrained for input data sampled at the Nyquist frequency in combination with spline interpolation show comparable results to that obtained with the reference NN model. Deconvolved profiles using the reference NN model show artefacts related to the noise if the input data has lower SNR than the original data used during the training. However, the NN models retrained using data with the same SNR as the input data produced denoised and deconvolved profiles with good agreement to the reference profiles. Further tests revealed that the same agreement can be achieved if the SNR of the training data is lower than that of the input data. Conclusion Retraining robust NN models demonstrated the same performance as the reference NN model using high resolution and high SNR training and input data. Even though the sampling rate has been decreased by a factor of 6 and the SNR reduced by a factor of 2, results of similar quality were achieved. The application of these robust NN models allows for faster beam profile scanning and correction for the detector volume effect, while preserving fine spatial resolution and high SNR of the resulting profiles. 1 Laboratory of Radiation Physics, Department of Oncology, Odense University Hospital, Odense, Denmark; 2 Department of Oncology, Odense University Hospital, Odense, Denmark Purpose or Objective PlanIQ TM (Sun Nuclear) software provides a tool that uses target and organ at risk (OAR) geometry to indicate the difficulty of achieving different doses for organ dose-volume histograms. We investigate whether this tool can be used as a priori estimation of the complexity of the plan. Is it possible to fulfill both goals for covering the target while sparing the OAR or is it necessary to compromise? This would be helpful upfront for the planner and oncologist to decide what to prioritize. Materials and Methods 108 breast cancer patients treated with postoperative radiotherapy at Odense University Hospital during 2020 were planned in Pinnacle 3 (version 16.2.1) and PlanIQ (version 2.2). The cohort of patients included both patients with lumpectomy, mastectomy, and with and without lymph node involvement. A plan was made in Pinnacle where all goals for the target (using DBCG consensus guidelines) were fulfilled while sparing the OAR as much as possible. Plan setup was a tangential field-in-field method. All plans were sent to PlanIQ including the target and the OAR goals and PlanIQ estimated the feasibility of fulfilling the OAR criteria while still fulfilling the target criteria. 54 of the patients were randomly selected and used as training set in order to fit a linear model relating PlanIQ doses (“predicted”) to Pinnacle doses (“actual”). The remaining 54 patients were used as a test set for validation of the fitted models. PO-1670 Prediction of heart and lung dose in breast cancer radiotherapy K.L. Gottlieb 1 , M. Kjellgren 1 , M.H. Nielsen 2 , K.H. Engstrøm 1 , E.L. Lorenzen 1

Results The linear model fitted on the training data was: µ HeartDose =1.270 ‧ µ PlanIQ +0.577 and µ LungDose =2.764 ‧ µ PlanIQ +0.166

µ PlanIQ being the predicted dose from PlanIQ and µ HeartDose and µ LungDose the mean heart and lung dose from Pinnacle dose plans for optimal target coverage. The result of applying a linear model to the training set is shown in Figure 1. a) and b). These plots show a scatterplot of the actual dose versus the predicted dose for the mean heart dose and mean ipsilateral lung