ESTRO 2024 - Abstract Book

S3103

Physics - Autosegmentation

ESTRO 2024

As for the pathological validation, the model tends to overestimate the tumor volume more than the manual delineations, with a median (range) volume overestimation of 5.2 (-4.0-17.5) mL and PPV of 0.52 (0.25-0.85) compared to 4.3 (-2.8-18.0) mL and 0.61 (0.36-0.83) for manual delineations. The model and the manual delineations achieved a similar tumor coverage with a median (range) sensitivity of 0.90 (0.60-0.99) and 0.91 (0.53-1.00), respectively. The margin to ensure coverage of at least 95% of the tumor surface in all patients was 6.4 mm and 6.6 mm for automatic and manual delineations, respectively. Example cases of the pathological validation are shown in Figure 1.

In the qualitative evaluation, automatic delineations were considered acceptable for treatment planning in 67% of cases, compared to 63% for manual delineations, demonstrating their potential for clinical utility.

Figure 1. Two example cases from the pathology dataset with the pathology (green), manual (blue) and automatic (red) delineations on sagittal slices. The first case (A) shows a good tumor coverage but also a large volume overestimation. In the second case (B), a large margin is needed to achieve adequate tumor coverage.

Conclusion:

We demonstrated that an MRI-only deep learning model for automatic laryngeal and hypopharyngeal GTV delineation achieves comparable segmentation results as manual delineations when compared to pathological ground truth tumor segmentation. The automatic delineations are acceptable for treatment planning, showing the potential for more consistency and efficiency in the radiotherapy planning process.

Keywords: Laryngeal and hypopharyngeal GTV, MRI, pathology

References:

[1] H. Ligtenberg et al., “Modality-specific target definition for laryngeal and hypopharyngeal cancer on FDG-PET, CT and MRI,” Radiotherapy and Oncology, vol. 123, no. 1, pp. 63–70, 2017, doi: https://doi.org/10.1016/j.radonc.2017.02.005. [2] E. A. Jager et al., “Interobserver variation among pathologists for delineation of tumor on H&E-sections of laryngeal and hypopharyngeal carcinoma. How good is the gold standard?,” Acta Oncol (Madr), vol. 55, no. 3, pp. 391 – 395, 2016, doi: 10.3109/0284186X.2015.1049661.