ESTRO 2024 - Abstract Book

S899

Clinical - CNS

ESTRO 2024

was located in one or more vertebrae. Beta coefficients together with 95% confidence intervals (CI) and Wald testing was used to report about the statistical significance of these differences using an α significance level of 0.05. Assumptions of normality and heteroscedasticity of residuals were verified, together with the assumption of having no multicollinearity among the two above mentioned covariables tested within the model.

Results:

In the observed group of patients, the average DSC, HDmax, HD95, AVD, RVD and DCM were 0.673±0.108, 19.81±7.09 mm, 9.17±3.16 mm, 56.97±55.19 cc, 40.50±14.19 %, 4.75±3.98 mm. After extending the automatic segmented image by 1 mm these metrics improved respectively to 0.757±0.094, 19.71±7.07 mm, 8.64±3.39 mm, 29.14±39.22 cc, 19.18±18.22 %, 4.39±3.91 mm. When observing the paired differences between the DSC for the raw and extended delineation we observed a positive increase by on average 0.084±0.035 in the DSC metric resulting in a higher coverage. Sub dividing this difference for the region and the covariable whether or not more vertebrae needed to be delineated this average increase was lessened to 0.059 (95% CI: 0.047-0.072). Statistically significant relations were found with the region in which the lesion was located (p<0.001) and with the covariable stating whether or not the lesion was located in one or more vertebrae (p=0.042). The gain in DSC in the cervical vertebrae region was 0.042 extra on average (95% CI: 0.025-0.058; p<0.001) compared to the lumbar region, by extending the automatic delineated CTV by 1mm. Also the gain in DCM in the thoracic region compared to the lumbar region (0.015; 95% CI: 0.000-0.030) was statistically significant (p=0.045). When contouring the vertebrae with the ESS software compared to the radiation oncologist, also an association was found between delineating multiple vertebrae and delineating only one vertebra. By adding a 1mm extra margin here, an average increase in DSC of 0.013 (95% CI: 0.000-0.026; p=0.042) was observed. The autosegmented CTVs generated by the ESS module for spine SBRT had a rather good similarity to those manually drawn by a radiation oncologist, which suggests that the ESS system could be a useful tool for generating CTVs in clinical practice. Additionally, adding some margin to the automatic delineation of the ESS module increases the overlap between them significantly. However, the increase was not always equally high for different regions among the spine and for lesions located in one vertebra compared to multiple vertebrae. Further research is necessary to find the most optimal margins to apply to a certain lesion. At this point, the software module could be easily used in clinical practice in a semi-automatic manner by projecting an initial delineation before continuing manual delineation of the radiation oncologist. Conclusion:

Keywords: Spine SBRT, automatic segmentation, delineation

2294

Digital Poster

Outcomes of multifocal glioma treated with radiation therapy: a single institutional study