ESTRO 2023 - Abstract Book

S1629

Digital Posters

ESTRO 2023

Materials and Methods From 24.01.2022 to 02.10.2022, 31 patients received prostate SBRT using flattening filter-free (FFF) volumetric modulated arc therapy (VMAT) for a primary (n = 27) or a intraprostatic recurrent PCa (n = 4). Patients were simulated and treated with a filled bladder (150cc) and an empty rectum. The prescription doses were 36.25Gy delivered in 5 consecutive fractions on the whole prostate, and 40Gy on the nodule(s) visible at the MRI, delivered with a Simultaneous Integrated Boost approach. The CTV-to-PTV margin was 2 mm for both the identified treatment volumes. Setup accuracy was verified with a ConeBeam-CT (CBCT) matching before each fraction. Noteworthy, in order to reduce the overall treatment time, we tested and adopted a fast CBCT (total duration of acquisition of the CBCT: 1 minute). The EM gating device consisted of a Foley catheter with an integrated EM transmitter. This transmitter is connected to the treatment table, thus allowing an online monitoring of the prostate displacements during the whole procedure (CBCT acquisition + treatment delivery). Treatment was manually interrupted when the signals exceeded a 2 mm threshold in any of the three spatial directions. A new CBCT was performed if the offset was transient (>20 secs.). Median and mean treatment times, calculated per fraction, were 10m31s and 12m44s (range: 6m36s - 65m28s), and 95% of the fractions were delivered with a maximum time of 27m48s. After the beginning of the delivery, the mean and median number of CBCT realized during the treatment were 2 and 1 (range: 0 - 11). During the treatment, the prostate was outside the CTV-to-PTV margin (2mm) for 11.2%, 8.9% and 3.9% of the treatment time, in the vertical, longitudinal and lateral direction, respectively, thus needing to stop the delivery +/- a reacquisition of the CBCT. Conclusion EM trasmitter-based gating for prostate SBRT is feasible . This technique could be quite easily integrated in the normal workflow of a Radiation Oncology Department, without any major impact on the agenda of the machines. Our results showed that, using this system, a 2mm CTV-to-PTV margin could be safetly applied. We found a small number of fractions with motion exceeding the predefined 2mm threshold, which would have otherwise gone undetected without intrafraction motion management. 1 Computational Neuroscience, University Medical Center Hamburg-Eppendorf, Hamburg, Germany; 2 Radiotherapy and Radiation Oncology, University Medical Center Hamburg-Eppendorf, Hamburg, Germany; 3 Siemens Healthcare, /, Erlangen, Germany Purpose or Objective Available motion management systems account for tumor motion with approaches such as tracking or gating to reduce normal tissue dose. However, the systems have to compensate for latencies using respiratory motion predictors. Present motion predictor comparison studies work on denoised breathing signals with a fixed short-term prediction horizon (usually 400-500ms). To potentially avoid required additional latencies for signal denoising, we extend the studies to noisy input signals, considering state-of-the-art prediction approaches: linear regression (LR) and long-short term memory (LSTM). In addition, we analyze the performance for longer prediction horizons required for other 4D SBRT applications (eg, CT and CBCT gating). Materials and Methods The study is based on 2517 respiratory signals (sampling rate 25Hz; length of signals between 30 and 300s) of 418 lung and liver SBRT patients, recorded with the Varian RPM system during planning 4D CT, pre-treatment 4D CBCT and gated dose delivery. Signals were randomly divided into training (number of patients n=215, number of total signals # 1265), validation (n=85, # 516) and test (n=118, # 736) set on patient level. Each signal was rescaled to an amplitude range [-1, 1], based on min/max scaling of its first 20s. To standardize noise conditions, each signal was Fourier-smoothed (1 Hz cutoff frequency), and white noise (SNR=27dB) was added afterwards. Inputs were sampled by an overlapping sliding window approach. Based on the noisy signal, the task was to predict the respiratory amplitude for the prediction horizon ahead (single output value prediction). Target signals were the denoised signal to avoid overfitting the models to the noise. The training loss was the mean squared error (MSE) between model output and the denoised target signal. Prediction horizons were 480ms (similar to literature), 760ms, and 1000ms. Results A representative case is shown in Figure 1. LR and LSTM test set prediction errors are summarized in Table 1. In line with current literature, the LR model performs almost perfectly on the denoised data and the short prediction horizon. However, for longer prediction horizons and noisy input, the prediction accuracy drops. Here, the LSTM model outperforms the LR model. For a 1s prediction horizon, the LSTM with noisy input performs on par with LR on denoised input data. Results We report data about 155 fractions. PO-1886 Respiratory motion prediction based on LSTM and linear regression models L. Wimmert 1 , M. Nielsen 1 , T. Gauer 2 , C. Hofmann 3 , R. Werner 1