ESTRO 2023 - Abstract Book

S1332

Digital Posters

ESTRO 2023

evaluation was carried out with a likert scale (0: accepted, 1: accepted with minor modifications, 2: accepted with major modifications, 3: rejected). Results Bladder delineation shows better results, while uterus and rectum have acceptable values despite a greater dispersion. The most important differences are observed for bowel bag which is the only structure obtained by DIR (Table 1). Greater dispersion is observed for patient 1 on bladder and bowel bag due to CBCT image quality and presence of gas (Figure 1). The dosimetric impact of using corrected or uncorrected contours is highlighted in Table 2 and is the most remarkable for bowel bag.

Dosimetric criteria are all respected for bladder and rectum whether the corrected or uncorrected contours are used. For bowel bag, however, it appears necessary to adjust the contours manually (online or offline). The qualitative evaluation shows that none of the contours provided by ETHOS were rejected, the majority were adjusted. Conclusion The evaluation shows excellent results for a majority of AI contours with CBCT images. The impact of manual correction on ETHOS contours is significant in terms of improvement of metrics, but the dosimetric impact is only significant for bowel bag, implying the need to recontour this structure during (or after) the oART CBCT-based session. Additional work is underway to optimize CBCT image quality.

PO-1636 Comprehensive analysis of different commercial auto-segmentation tools for multi-site OAR contouring

G. Heilemann 1 , M. Buschmann 1 , W. Lechner 1 , V.M. Dick 1 , F. Eckert 1 , M. Heilmann 1 , H. Herrmann 1 , M. Hudelist-Venz 1 , M. Moll 1 , J. Knoth 1 , S. Konrad 1 , I.M. Simek 1 , C. Thiele 1 , Z. Alexandru-Teodor 1 , D. Georg 1 , J. Widder 1 , T. Petra 1

1 Medical University of Vienna, Department of Radiation Oncology, Vienna, Austria

Purpose or Objective To analyze the performance of commercial organs-at-risk (OAR) auto-segmentation software for clinical use and to describe the methodology for clinical validation. Materials and Methods AI-based contours were automatically generated by two commercial software tools on CT images for representative sets of at least 15 patients for brain, head-and-neck, thorax, abdomen, as well as female and male pelvis, and benchmarked against respective manual experts’ contours. Following metrics were statistically analyzed for more than 25 different OARs: 1) geometric (Dice similarity coefficient (DSC), Hausdorff distance (HD)); 2) dosimetric (relative dose differences of mean and near maximum dose D1%); 3) qualitative metrics: physician rating from 1 = poor quality to 4 = good quality/clinically acceptable, rated at least by two physicians per structure. Results Contours of a total of 180 patients were evaluated with respect to more than 25 different OARs. The geometric evaluation yielded good results with a mean DSC above 0.9 for 29% of all OARS and 59% above 0.8. Few structures resulted in a poor