ESTRO 2021 Abstract Book

S1380

ESTRO 2021

Italy; 4 Florentine Institute of Care and Assistance (IFCA), Radiation Oncology Unit, Florence, Italy; 5 University of Florence, Radiation Oncology Unit, Florence, Italy; 6 University of Florence, Medical Physics Unit, Florence, Italy Purpose or Objective The aim of this work is to evaluate the introduction of an auto-segmentation tool, for pelvis CT images, into the clinical workflow and to evaluate its impact on contouring time, intra and inter observer variability (intra- OV and inter-OV) and contours accuracy. Materials and Methods An atlas for male pelvis CT auto-segmentation, created and optimized in a previous work [1], was used to automatically contour pelvic-lymph-nodes, rectum, bladder and femurs of 6 test subjects. 5 radiation oncologists (ROs) manually delineated the contours (M1) and corrected the automatically generated contours (AMC1). After a minimum period of 3 weeks, the same procedure was repeated generating M2 and AMC2. Contouring / editing time were registered and compared to evaluate possible time saving. Intra-OV was assessed by comparing M2 to M1 and AMC2 to AMC1 (fig. 1A) by means of Dice Similarity Coefficient (DSC) and Mean Distance to Agreement (MDA). Obtained values relative to AMC were paired and compared to manual values in a Wilcoxon signed rank test (both for DSC and MDA). Regarding inter-OV, for each ROI and each test patient, M1 contours by the 5 ROs were combined with STAPLE finalization algorithm to obtain a common reference Ref_M1, analogously for AMC1 contours (Ref_AMC1). Inter-OV was assessed, both for M1 and AMC1 contours, by comparing each RO’s contour to the common reference and quantified with DSC and MDA indices (fig. 1B). DSC and MDA values for AMC contours were compared to those for manual contours, applying a paired Wilcoxon signed rank test.

Results A considerable time sparing of 45% was achieved, averaged on ROs and ROIs. Percentage time sparing varies with ROs (34% to 55%, 12 min to 23 min whole pelvis) and ROIs (1.3 min, 32% rectum to 6.4min, 74% femurs, average on ROs). Both for DSC and MDA, a significant (p<0.001) reduction of inter-OV was observed (fig. 2B DSC), comparing AMC to manual approach, for all the structures. As for intra-OV (fig. 2 A), the reduction was significative only for bladder and femurs.