ESTRO 2022 - Abstract Book

S1496

Abstract book

ESTRO 2022

1 Amsterdam UMC - VUmc location, de Boelelaan 1117, 1081 HV, Department of Radiation Oncology, Amsterdam, The Netherlands; 2 Viewray Inc., 2 Thermo Fischer Way, Science Department, Oakwood Village, Ohio 44146, USA Purpose or Objective MR-guided radiotherapy (MRgRT) enables real-time monitoring of the anatomy of the patient during treatment delivery. Manual delineations for tumor contour tracking during delivery is impractical and very labor intensive. In this work we studied the accuracy of automatic tumor segmentation based on deformable image registration (DIR) in the delivery of MRgRT by comparing it to manual delineations performed by experienced observers. Materials and Methods Twenty patients with lung, pancreatic, renal and adrenal tumors previously treated with SBRT using MRgRT were included in the study. Patient and treatment fractions included for this analysis were randomly selected. Five observers with at least 2 years of experience in MRgRT delineated the gross tumor volume (GTV) for 20 patients on 240 frames of an MR-cine on a sagittal plane, with 0.35cm x 0.35cm in-plane resolution. DIR-based GTV contours were propagated using 4 different algorithms from a reference frame to subsequent frames, in order to assess the accuracy of online DIR-based contour tracking. Each DIR-algorithm was implemented as a combination of several independent tracking modules, referred to as “trackers”. Geometrical analysis based on the Dice Similarity Coefficient (DSC), centroid distance and Hausdorff Distance (HDD) were performed to assess the inter-observer variability and the accuracy of automatic segmentation. A Confidence Value metric for the reliability of the tumor auto-contouring was also calculated. The Confidence Value is based on the correlation of the image registration and intersection and union of the results from each of the trackers composing the algorithms. Results Figure 1A shows the delineations of the experts for an adrenal and lung case. Inter-observer delineation variability resulted in mean DSC, HDD and centroid distance among observers across all patients of 0.89, 5.8 mm and 1.7 mm, respectively. Figure 1B shows the comparison for the DSC values between the experts and one of the algorithms for all tumor sites. The four DIR algorithms were able to reproduce the original contour delineated by each of the observers on the reference frame to subsequent frames with an excellent agreement. Mean DSC for each algorithm across all patients and frames was > 0.90, whereas the HDD and centroid distances were below 4.0 mm and 1.5 mm, respectively. All four algorithms performed equally good independently of the specific tumor site. The Confidence Value was linearly correlated with the DSC.

Conclusion DIR-based auto-contouring in MRgRT exhibited a high level of agreement with the manual contouring performed by experts. In addition, DIR-based auto-contouring resulted in a lower variability across all quantitative metrics compared to the inter- observer variability. Clinical implementation of DIR-based algorithms for intra-fraction motion correction in MRgRT provides clinicians and physicists with robust and reliable tools to deliver radiation dose accurately.

PO-1696 Comparison of intra-fraction motion when using two distinct systems to perform DIBH in liver SBRT.