SXX
Corrigendum
ESTRO 2026
of edited slices (%EDIT), Volumetric DSC (VDSC), and Percentage volume difference (%dVOL) using a research version of AIQUALIS (Inpictura Ltd, Abingdon, UK).
Conclusion: Non-linear measures (e.g. NAPL, SDSC, and %EDIT ) are more sensitive to the type of automation bias simulated than linear measure (e.g. MSD, VDSC). These measure may be the most appropriate to detect automation bias in clinical practice. Nevertheless, further research is required to understand how automation bias manifests itself in the clinical editing process. References: [1] Nealon, K. et al. (2024). Monitoring variations in the use of automated contouring software. Practical radiation oncology, 14(1), e75-e85. [2] van der Velden, S. et al. (2025). Assessing the long- term clinical usage of auto-segmentation for head and neck organs-at-risk. Radiother Oncol, 206 , S2491-S2492. [3] Doolan, P. et al. (2025) How to spot, and stop, automation bias in autocontouring. Radiother Oncol, 206 , S2468-S2469 [4] Rong, Y. at al. (2025) Quality monitoring of AI auto- segmentation performance for prostate cancer patients. Radiother Oncol, 206 , S2519-2520 [5]. Nikolov, S et al. (2018) Deep learning to achieve clinically applicable segmentation of head and neck anatomy for radiotherapy, https://arxiv.org/abs/1809.04430 Keywords: statistical process control, contouring, metrics
Results: For LED-bias simulation, NAPL was most sensitive, detecting automation bias after 4, 9 and 10 cases (after bias began) for Parotid_L, Brainstem and Mandible respectively. SDSC was similarly sensitive. %dVol and 3D-95%-HD were found to be least sensitive. For the IOU-bias simulation, %EDIT was most sensitive, detecting bias after 24, 7 and 36 respectively for the same organs. NAPL and SDSC were also quite sensitive. Again %dVol and 3D-95%-HD were least sensitive, with %dVol failing to detect any bias for the Parotid_L for this simulation. Figure 2 illustrates the measures’ absolute z-scores plotted against the cases for the Brainstem using the LED-bias simulation and for the Parotid_L using the IOU-bias simulation.
Made with FlippingBook - Share PDF online