ESTRO 2024 - Abstract Book

S3091

Physics - Autosegmentation

ESTRO 2024

3. Choi M. S. et al. Radiotherapy and Oncology 153 (Dec. 2020), p. 139–145. doi: 10.1016/ j.radonc.2020.09.045

4. Chung S. Y. et al. Radiation Oncology 16.1 (Feb. 2021). doi: 10.1186/s13014-021- 01771-z

5. Liu Z. et al. Cancer Management and Research Volume 13 (Nov. 2021), p. 8209–8217. doi: 10.2147/cmar.s330249

6. Almberg S. S. et al. Radiotherapy and Oncology 173 (Aug. 2022), p. 62–68. doi: 10.1016/j.radonc.2022.05.018

7. Bell L. R. et al. Journal of Physics: Conference Series 777 (Jan. 2017), p. 012002. doi:10.1088/1742 6596/777/1/012002

2136

Digital Poster

Leveraging AI autosegmentation to assess carotid stenosis risk after definitive H&N radiotherapy

Daniel K Ebner, Jing Qian, Jack C Thull, Connie Geno, Daniel J Ma

Mayo Clinic, Radiation Oncology, Rochester, USA

Purpose/Objective:

Radiotherapy (RT)-induced carotid stenosis for head and neck cancer patients is of growing concern [1], and currently no good constraint or high-risk guidelines exist for patients. Traditional stenosis diagnoses, performed via ultrasound or angiography [2-4] and five to ten years post-RT, do not capture or allow intervention upon early-onset cases. This study analyzes time series contrast-free CT scans conducted before and after definitive RT, leveraging deep learning autosegmentation on simulation and diagnostic scans to identify correlations with eventual diagnosis of carotid stenosis on ultrasound. This study is a methodological pilot aiming to pinpoint image features associated with carotid stenosis, identifying patients at high risk for treatment planning considerations, early follow-up, and evaluation.

Material/Methods:

Patients with oropharyngeal cancer treated with definitive photon or proton radiotherapy consisting of 70 Gy in 33 35 fractions between 2013 and 2021, who had received carotid ultrasounds following their radiotherapy, were identified. Those with prior or concurrent carotid stenosis diagnosis were excluded, as were patients with multiple courses of radiotherapy to the same area of the head or neck. These criteria resulted in 62 patients being selected. Simulation, diagnostic, and followup PET/CT and CT scans were identified, with bilateral carotids segmented using a deep learning algorithm (MIM ProtegeAI). The higher dose level was considered ipsilateral. Carotid calcification ratios were estimated by analyzing the percentage of high Hounsfield Unit (HU) voxel volumes within the carotids in range of the oral cavity. Carotid size was estimated by using the mean cross-sectional area to account for possible variance in AI contouring. Timepoints at simulation, 4 months, and 1 year after RT for patients with and without stenosis were trialed, respectively.

Results: