Session Item

Accurate segmentation of organs at risk (OARs) and target volumes is crucial for radiation treatment planning but highly time-consuming. One commercially available solution is atlas-based auto-segmentation (ABAS, Elekta, Stockholm). Its use aims to reduce time-consumption and inter- and intra-observer’s variability, which may significantly affect dosimetric parameters. The purpose of this study was to investigate the use of ABAS in gynecological clinical routine.

This feasibility study retrospectively selected 23 patients treated with external beam radiation therapy (EBRT) between 2019 and 2021. Ten out of 23 structure sets were used to create an atlas library in ABAS containing bladder, rectum, and femoral heads carefully defined by two skilled radiation oncologists. The remaining 13 patient CTs were used to test the library performances. The following contour comparison was performed using the Hausdorff distance (HD), the Mean Distance to Agreement (MDA), the Dice similarity Index (DI), and the Jaccard Index (JI). The results were compared performing the Wilcoxon Mann Whitney test to assess statistical significance (α=0.05).

The auto-segmentation process required about 6 minutes for each CT set. Figures 1 and 2 show the geometrical metric results. The median HD, MDA, DI, and JI values for the right femoral head were 7.44 mm [3.66 – 17.77 mm], 1.20 mm [0.67 – 2.12 mm], 0.89 [0.80 – 0.93], and 0.80 [0.67 – 0.88], retrospectively; for the left femoral head were 7.53 mm [3.87 – 15.54 mm], 1.13 mm [0.74 – 1.39 mm], 0.91 [0.86 – 0.93], and 0.84 [0.76 – 0.87]. On the other hand, median metric values for the bladder were 20.64 mm [12.00 – 34.95 mm], 3.13 mm [1.93 – 11.21 mm], 0.80 [0.36 – 0.87], and 0.66 [0.22 – 0.77], respectively. Lastly, the median metric results for the rectum were 23.79 mm [14.81 – 42.56 mm], 3.58 mm [2.19 – 9.98 mm], 0.64 [0.34 – 0.81], and 0.47 [0.21 – 0.68], respectively. The data comparison showed comparable ABAS performances in rectum and bladder segmentation (p > 0.05) and in right and left femoral heads (p > 0.05). A significant difference was registered for ABAS segmentation performances in bilateral-femoral heads compared to rectum and bladder (p << 0.05).

In these results, ABAS-bilateral-femoral heads segmentation showed higher performance compared to ABAS-bladder and ABAS-rectum segmentation because the significant bone density interface could help ABAS in femoral head delineation. Although in some cases the auto-segmentation still required manual corrections, its implementation in daily clinical practice could change the physician workflow increasing the consistency and the time-saving.