ESTRO 2020 Abstract book

S1013 ESTRO 2020

require the acquisition of the entire body making HDFoV a necessity. This work evaluates HDFoV images obtained with two different Siemens software versions, VA62 and VB10. Material and Methods CT raw data was acquired for various patients (head and neck, thorax and pelvis) and phantoms. CT scans were reconstructed with two software versions: VA62 and a newer algorithm VB10. Phantom acquisitions (Gammex and Kyoto phantoms) were performed at the scanning isocenter and laterally shifted so part of the phantom was outside the standard 50 cm reconstruction diameter. In addition, a 3D printed phantom based on the anatomy of a breast cancer patient, for whom a HDFoV acquisition was necessary, was manufactured. This 3D printed phantom has holes for tissue mimicking Gammex inserts (used to verify HU for a range of materials - Table 1), a low-density material mimicking lung tissue, and PLA plastic to mimic soft tissues. A treatment plan with one 6MV photon beam was created for the 3D printed phantom to assess the dose distribution differences.

PO-1733 Validation of PSMA-PET/CT based contouring techniques for intraprostatic tumor definition D. Kostyszyn 1 , T. Fechter 1 , D. Baltas 1 , J. Ruf 2 , M. Mix 2 , P. Bronsert 3 , A. Grosu 1 , C. Zamboglou 1 1 University Medical Centre Freiburg, Department of Radiation Oncology, Freiburg im Breisgau, Germany ; 2 University Medical Centre Freiburg, Department of Nuclear Medicine, Freiburg im Breisgau, Germany ; 3 University Medical Centre Freiburg, Institue for Surgical Pathology, Freiburg im Breisgau, Germany Purpose or Objective Defining the intraprostatic gross tumor volume (GTV) on prostate specific membrane antigen positron emission tomography (PSMA-PET) is crucial for diagnosis and therapy in the course of focal radiation therapy. So far, no optimal contouring method could be defined. This study presents a profound analysis of manual and automatic 20 prostate cancer (PCa) patients were enrolled. Each underwent a [ 68 Ga]PSMA-11-PET/CT followed by radical prostatectomy. Six observer teams with different levels of experience performed manual contouring of GTV, using different PSMA-PET standard uptake value (SUV) image scaling techniques. Coregistered histopathological gross tumor volume (GTV-Histo) served as reference for validation. Subsequently, the manual segmentation technique with highest sensitivity, which was stated as best performing technique, was used to train a deep convolutional neural network (CNN) with another 64 patients (GTV-CNNs). GTV-CNNs were assessed by the Dice similarity coefficient (DSC) with the best manual contours. Results In slice by slice comparison with histology information, most manual contouring methods provided high sensitivity (32-100%) and specificity (43-100%). Scaling the PSMA-PET images from SUVmin-max: 0–5 resulted in highest sensitivity (>86%). High interobserver agreement (median DSC 0.8) was observed by three teams with different experience when using the same PSMA-PET image scaling technique (SUVmin-max: 0–5). The highest median specificity (100%) was obtained by scaling the PSMA-PET images in full range: 0-SUVmax. The CNN provided a DSC of 0.76 in comparison with the best performing manual segmentation. Computation time was between three and four seconds per patient. Conclusion Manual contouring using validated SUV scaling may provide high sensitivity (SUVmin-max: 0–5) or high specificity (SUV full range). Automatic contouring with a CNN provided human expert level accuracy. Both methods could be considered for PSMA-PET-based intraprostatic GTV- delineation. Due to the fast contouring, the CNN method may be considered as a supporting tool for manual GTV- delineation in order to decrease treatment planning time. PO-1734 Assessment of extended field-of-view CT reconstructions for radiotherapy G. Fonseca 1 , B. Van der Heyden 1 , I. Almeida 2 , I. Rinaldi 1 , W. Van Elmpt 1 , F. Verhaegen 1 1 Maastricht University Medical Centre+, Department of Radiation Oncology MAASTRO, Maastricht, The Netherlands ; 2 Technical Universty of Lisbon, Center of Nuclear Sciences and Technologies, Lisboa, Portugal Purpose or Objective Siemens CT scanners can reconstruct images up to the bore size (typically 80 cm on Siemens large bore scanners). Truncated data is used to reconstruct approximated images outside the standard scan-field-of-view (sFOV – typically 50 cm). The extended field-of-view (HDFoV – High definition field of View Pro) feature has been used for patients with higher BMI or non-isocentric positioning due to fixation devices. Accurate radiation dose calculations contouring approaches. Material and Methods

Results Figure 1a shows a CT slice reconstructed using VA62 and body contours obtained with sFOV and HDFoV. Differences depend on the body region with a maximum deviation of 9 mm between both algorithms. VB10 showed a better performance than VA62 for Gammex and Kyoto phantoms with body contour deviations approximately 2-fold lower than VA62. Figure 1b shows the 3D printed phantom with Gammex inserts. HDFoV (VB10), sFOV and reference body contours are shown in Figure 1c. Body contour deviations are generally below 5 mm and HU deviations reached a maximum of -368 HU for cortical bone, 257 HU for lung and 102 HU for adipose tissue (Table 1). Dose deviations show a constant offset of -1.6% (Figure 1d) in the region where HDFoV and reference body contours overlap.