ESTRO 2021 Abstract Book


ESTRO 2021

plans (worst-case vs. probabilistic). Both methods attain similar levels of target coverage, both in terms of worst-case D95 and robustness (indicated by the DVH bandwidth (BW)). The probabilistic method shows a slight increase in mean lung dose whilst either being equivalent or improving the esophagus D2 dose.

Conclusion A fully probabilistic approach is implemented, which evaluates a probabilistic target during a robust optimization process. The probabilistic approach allows the optimizer to redefine compromises and hence, achieve promising results in terms of the trade-off between target coverage and OAR sparing. PH-0042 Dosimetric impact of auto segmentation on treatment planning in IMRT for prostate patients M. Kawula 1 , D. Purice 1,2 , M. Li 3 , G. Vivar 4 , S. Ahmadi 5 , K. Parodi 6 , C. Belka 3,7 , G. Landry 3,8 , C. Kurz 3,8 1 University Hospital, LMU Munich, Department of Radiation Oncology, Munich, Germany; 2 Ludwig- Maximilians-Universität München, Department of Medical Physics, Garching, Germany; 3 University Hospital, LMU Munich, Department of Radiation Oncology , Munich, Germany; 4 Ludwig-Maximilians-Universität München, German Center for Vertigo and Balance Disorders, Planegg, Germany; 5 Ludwig-Maximilians- Universität München, German Center for Vertigo and Balance Disorders , Planegg, Germany; 6 Ludwig- Maximilians-Universität München, Department of Medical Physics, Garching, Germany; 7 German Cancer Consortium, (DKTK), Munich, Germany; 8 Ludwig-Maximilians-Universität München, Department of Medical Physics , Garching, Germany Purpose or Objective Accurate automatic delineation of organs and structures of interest can assist online workflows in intensity- modulated radiation therapy (IMRT), supporting or replacing manual segmentation, which is a time-consuming process prone to inter- and intra-observer variances. Besides a direct comparison of the auto-generated segmentations against expert segmentations, e.g. by dice similarity coefficient (DSC) and Hausdorff distance (HD), a subsequent evaluation based on dosimetric parameters carries a higher relevance in clinical practice. In this study, we investigated the impact of state-of-the-art 3D . U-Net-generated organ delineation for computed tomography (CT) images on IMRT dose optimization in the male pelvic region. Materials and Methods The dataset, consisting of 69 CT scans along with prostate, bladder, and rectum segmentations, was subdivided into 45 training, 11 validation, and 13 test cases. Data augmentation through rotations, translations, and deformations was used to supplement the training set. A 3D U-Net architecture with DSC loss was trained separately for each organ. For all test cases, volumetric-modulated arc (VMAT) treatment plans were generated with the same optimization settings for both manual and automatic segmentation. Care was taken to find optimization settings producing consistent results by applying small perturbations to the manual segmentations. A 6 mm PTV margin (posterior 5 mm) around the prostate was applied. Dose distributions were evaluated with the manual segmentation as reference using dose-volume histogram (DVH) parameters and a 3%/3mm gamma-criterion with a 10% dose threshold, in addition to DSC and HD analysis. For DVH analysis, the prostate (with and without 3mm margin) D 98/2% and V 95% , were considered. For rectum V 50/65/70 Gy and bladder V 60/65/70 Gy were analyzed. Results Figure 1 shows exemplary image slices of two patients together with the corresponding dose distributions optimized on manually and automatically generated contours. On average the gamma pass rates were 85.1% for the 11 test cases. The results of the DVH analysis are presented in Figure 2. With the exception of one case, a satisfactory agreement between the two dose distributions was observed. Contours were evaluated in terms of DSC, average and 95% HD demonstrating a performance of DSC prostate = 0.871(0.001), DSC bladder = 0.965(0.001), DSC rectum = 0.888(0.001) and HD avg /HD 95%|prostate = 1.6(0.2)/4.2(1.1) mm, HD avg /HD 95%|rectum = 1.4(0.4)/5(12) mm, HD avg /HD 95%|bladder =0.95(0.03)/2.5(0.3) mm.

Made with FlippingBook Learn more on our blog