Related Experiment Video
Updated: Jul 2, 2026
Irradiator Commissioning and Dosimetry for Assessment of LQ α and β Parameters, Radiation Dosing Schema, and in vivo Dose Deposition
Published on: March 11, 2021
Radiotherapy quality assurance in the PRO-GLIO trial: results from a dummy run comparing experts across twelve
Liv Cathrine Heggebø1,2, Taran Paulsen Hellebust3,4, Thomas Henry5,6
1Department of Oncology, Oslo University Hospital, Oslo, Norway.
Background:
Target volume and organ of interest (OOI) contouring and treatment planning for IDH-mutated gliomas CNS WHO grades 2 and 3 are subject to interindividual variation. We aimed to assess this variation and heighten delineation awareness as a quality assurance measure in the PRO-GLIO trial.
Methods:
Five experts established consensus delineations for target volumes and defined OOI in two IDH-mutated gliomas cases. Next, target volumes and OOI were delineated by experts from 12 treatment centers participating in the PRO-GLIO trial, and by deep learning segmentation (DLS). These structures were compared to consensus contours based on predefined quantitative and qualitative parameters. Proton and photon treatment planning were performed on the consensus delineations by 11 centers.
Results:
Median dice similarity coefficient (DSC) for clinical target volume (CTV) was 0.92 for both cases, whereas the median 95th percentile Hausdorff distance for CTV was 0.54 cm and 0.65 cm. Larger variability in DSC and volume size was seen for OOI, e.g., a median DSC of 0.47 and 0.64 for the optic chiasm and between 0.61 and 0.77 for the hippocampi. DSC-derived DLS-values were within the range of manual delineations for all OOI.
Conclusions:
Delineation variation between study centers was minor for target volumes, especially for DSC of CTV. Variability was considerably higher for several OOI, representing a potential hazard not least with new and more precise radiotherapy techniques. Interestingly, DLS performed OOI-segmentation within the range of manual experts and represents a reasonable OOI-segmentation alternative in the future.
