ESTRO 2020 Abstract book

S993 ESTRO 2020

In a simulation study we tested the performance of the optimization algorithm using an anthropomorphic head phantom and a ROI corresponding to beam angles at 90° and 180°. The prescribed image variance was V ROI = 0.0012 (3.4% RSP standard deviation) inside the ROI and 4 ∙ V ROI elsewhere. We calculated image variance maps and evaluated imaging doses. For a fair comparison, results were normalized such that peak variance inside the ROI was equal. FMpCT scans were compared to standard scans with uniform fluence. Moreover, we present first experimental results acquired at a PBS beamline using a prototype pCT scanner and compare them to the simulations. Challenges and limitations of employing varying fluence fields are discussed. Results Figure 1 shows uniform fluence and FMpCT scans from the simulation study (a,b) as well as experimental scans (c,d). RSP and variance maps are shown, and for the uniform fluence scans (a,c), variance is elevated around heterogeneities. Variance in the FMpCT scans (b,d) follows the prescription while some streaks originating from heterogeneities distort the result. Profiles show that variance inside the ROI could be achieved. Mean RSP accuracy was retained in the FMpCT-ROI. Mean imaging dose for uniform fluence was 1.37 mGy. For the FMpCT scan it was 0.83 mGy outside the ROI and 1.65 mGy inside, and on average 1.11 mGy. Out-of-ROI dose reduction was 39%. Doses in the experiment could not be obtained, but are assumed to be comparable, as the number of protons per pixel (counts) are similar for experiment and simulation.

image. To improve the accuracy of the tumor segmentation, an attention boosted algorithm was implemented. Material and Methods Two-hundred-eighty-six parotid tumor patients were enrolled into this study. Two deep learning-based segmentation networks were developed, including a Unet and an attention boosted Unet (Figure 1). MRI scans (3T, T1 and T1 contrast enhanced) of each patient were collected for model training and valuation. Data were randomly separated into training (90%) and validation (10%) datasets. The Dice index (DSC) was calculated to compare the performance of the auto segmentation algorithm.

Figure 1. The network of attention boosted convolution neural network. Results The DSC was 0.81, 0.81, 0.89 and 0.84 for right-parotid, left-parotid, right-tumor and left-tumor by attention boosted convolution neural network, respectively. Meanwhile, the DSC was 0.82, 0.62, 0.85 and 0.80 for right- parotid, left-parotid, right-tumor and left-tumor by traditional Unet. Conclusion This study showed that attention boosted convolution neural network can perform better than Unet for parotid tumor and parotid segmentation on MRI images. PO-1706 Low dose fluence-modulated proton CT: simulation study and first experimental results J. Dickmann 1 , C. Sarosiek 2 , G. Coutrakon 2 , S. Rit 3 , N. Detrich 4,5 , V. Rykalin 6 , M. Pankuch 5 , R.P. Johnson 7 , R.W. Schulte 8 , K. Parodi 1 , G. Dedes 1 , G. Landry 1,9,10 1 Ludwig-Maximilians-Universität München, Department of Medical Physics, Garching bei München, Germany ; 2 Northern Illinois University, Department of Physics, DeKalb IL, USA ; 3 Univ Lyon, Creatis UMR 5220, Lyon, France ; 4 IBA International, Louvain-La-Neuve, Belgium ; 5 Northwestern Medicine Chicago Proton Center, Chicago IL, USA ; 6 Proton VDA Inc., Naperville IL, USA ; 7 University of California Santa Cruz, Department of Physics, Santa Cruz CA, USA ; 8 Loma Linda University, Division of Biomedical Engineering Sciences, Loma Linda CA, USA ; 9 University Hospital LMU Munich, Department of Radiation Oncology, Munich, Germany ; 10 German Cancer Consortium (DKTK) Munich, Germany, Purpose or Objective To use fluence-modulated proton CT (FMpCT) to achieve task-specific image quality. This can minimize imaging dose to obtain accurate relative stopping power (RSP) maps required for particle therapy treatment planning, and enable daily dose verification in treatment position. Material and Methods Employing task-specific proton fluence fields with pencil beam scanning (PBS) can reduce imaging dose while maintaining image quality inside the treatment beam path region-of-interest (ROI). A dose reduction is expected outside the ROI, which is desirable given particle therapy’s low integral out-of-field dose. We present an optimization algorithm that calculates relative pencil beam weights, which yield a prescribed image variance target. It is based on an object- and pCT-scanner-specific noise model obtained using a dedicated Monte Carlo simulation.

. Conclusion We obtained object- and task-specific fluence modulations using an optimization algorithm for FMpCT. A considerable dose saving of 39% compared to a uniform fluence scan could be achieved in a simulation study. The investigated shape of the ROI was relevant for clinical practice and the FMpCT image conveys all relevant information for planning, at equal RSP accuracy and reduced dose. We also demonstrated the experimental feasibility of FMpCT by applying the same fluence pattern in a scan using a prototype pCT scanner. Acknowledgement: ESTRO TTG, DFG project #388731804, BaCaTeC A1[2018-2] PO-1707 Fibroblast Activation Protein (FAPI) PET for diagnostics and radiotherapy in head and neck cancer M. Syed 1,2,3 , P. Flechsig 4 , J. Liermann 1,2,3 , J. Debus 1,2,3 , F. Giesel 4 , U. Haberkorn 4 , S. Adeberg 1,2,3 1 University Hospital Heidelberg, Department of Radiation Oncology, Heidelberg, Germany ; 2 University Hospital Heidelberg, Heidelberg Institute of Radiation Oncology HIRO, Heidelberg, Germany ; 3 University Hospital