Supplementary figures to Nielsen, C. P., Lorenzen, E. L., Jensen, K., Eriksen, J. G., Johansen, J., Gyldenkerne, N., Zukauskaite, R., Kjellgren, M., Maare, C., Lønkvist, C. K., Nowicka-Matus, K., Szejniuk, W. M., Farhadi, M., Ujmajuridze, Z., Marienhagen, K., Johansen, T. S., Friborg, J., Overgaard, J., & Hansen, C. R. (2024). Interobserver variation in organs at risk contouring in head and neck cancer according to the DAHANCA guidelines. Radiotherapy & Oncology, 110337. https://doi.org/10.1016/j.radonc.2024.110337
Please cite the above paper when reffering to these figures.
For questions, please contact the corresponding author Camilla Panduro Nielsen camilla.panduro.nielsen@rsyd.dk
Abstract
Background and purpose: There is variation in contouring of organs at risk (OARs) due to interpretation of anatomy and delineation guidelines, heterogeneity in image quality, and time pressure in the clinic. This study investigated interobserver variations among experts who contoured following national OAR guidelines.
Materials and Methods: Prior to a two-day workshop, experts clarified and agreed on the Danish Head- and Neck Cancer guidelines for OAR contouring. 15 expert oncologists and radiotherapy technologists contoured OARs on CT scans from 26 head- and neck cancer patients. Variation between contours was evaluated pairwise, regardless of number of assessors for each OAR, using concordance metrics like dice similarity coefficient (dice) and mean surface distance (MSD). Additionally investigating the 3D variation in specific areas of the OARs, the mean standard deviation (SD) on the contour surfaces of each OAR was calculated as a mean of all patients and observers for each OAR.
Results: 3545 OARs were delineated. The median and interquartile range number of observers per OAR was 9 [7 - 9]. The interobserver variation in contouring was organ specific with median dice ranging at 0.48 - 0.84 and median MSD ranging at 1.0 - 3.1 mm. Visualising mean SD highlighted specific areas of OARs with higher contouring variation for ill-defined organs and contour borders.
Conclusion: Knowing the interobserver variation of OARs emphasises where guidelines are less clear or where OARs are difficult to contour. This could drive the discussion to clarify clinical guidelines and could serve as a comparator for automated delineation systems.