S623
Clinical – Head & neck
ESTRO 2026
implementation cohort, masseterContra and masseterIpsi Dmean guideline adherence was 97% and 79 while being 99% for medial pterygoidContra and 93% for medial pterygoidIpsi. For each masseter and medial pterygoid, trismus was more prevalent if the Dmean guideline was not adhered to. A
Digital Poster Highlight 3386 Reducing masticatory muscle doses and patient- reported trismus via implemented AI auto- segmentation-enabled dose/volume mitigation Maria Thor 1 , Aditi Iyer 1 , Caroline E Olsson 2 , Aditya Apte 1 , Michalis Aristophanus 1 , Sharif Elguindi 1 , Jean Moran 1 , Nancy Y Lee 3 , Joseph O Deasy 1 1 Medical Physics, Memorial Sloan Kettering Cancer Center, NYC, USA. 2 Medical Radiation Sciences, Institute of Clinical Sciences, Sahlgrenska Academy, Gothenburg, Sweden. 3 Radiation Oncology, Memorial Sloan Kettering Cancer Center, NYC, USA Purpose/Objective: Trismus limits daily life in patients treated with radiotherapy (RT) for head-and-neck cancer (HNC). While mouth-opening ability is typically monitored after RT, methods to minimize the risk of radiation- induced trismus are scarce. This study implemented mean dose (Dmean) guidelines to artificial intelligence (AI) auto-segmented masticatory muscles and studied the impact on trismus after RT for HNC. Material/Methods: The masseters and medial pterygoids Dmean guidelines were clinically implemented in 2021 (masseter <42 Gy; medial pterygoid <68 Gy). All included HNC patients had been consecutively treated six months before (N=115) vs. six months after (N=110) the Dmean guidelines implementation. Trismus was patient-reported through the validated 21-item Gothenburg Trismus Questionnaire and defined as ≥ moderate from the maximum-recorded severity within one year after RT. Mastication muscles in the pre-implementation cohort were auto- segmented retrospectively using the clinical AI algorithm. The mastication-muscle-specific Dmean was compared between cohorts using the Wilcoxon rank-sum test (significance: p<0.05). In the post- implementation cohort and for trismus items with ≥ 15% prevalence, Dmean-response relationships were derived for the masseter or the medial pterygoid, which utilized logistic regression with bootstrapping (1000 samples) in which discrimination (AUC, p-values) and calibration (Hosmer-Lemeshow test p-values, pHL) were assessed as the median over samples. Results: Most patients had tumors in the oropharynx (pre- vs. post-implementation cohort: 50% vs. 56%), followed by in the oral cavity (14% vs. 16%) and were treated to a median of 60 (interquartile range (IQR); pre: 52-70; post: 44- 70) Gy. MasseterContra Dmean was significantly higher in the pre- vs. in the post- implementation cohort: median (IQR): 14 (6-20) Gy vs. 10 (6-14) Gy (p=0.01), and numerically higher for medial pterygoidContra (p=0.06), masseterIpsi and medial pterygoidIpsi (p=0.19). In the post-
≥ 15% prevalence (range: 15-37%) was observed for 15/21 trismus items of which 3/6/2/2/2 items were related to eating/jaw/pain/quality-of-life/teeth-tongue. Significant Dmean-response relationships were derived for all 15 items, and masseterIpsi provided the strongest relationship (15-item median: AUC=0.74, pHL=0.57) followed by masseterContra (AUC=0.71, pHL=0.65), medial pterygoidIpsi (AUC=0.69, pHL=0.64), and lastly medial pterygoidContra (AUC=0.67, pHL=0.62). Conclusion: This work demonstrates the feasibility of clinically implementing mastication-muscle-sparing dose/volume guidelines enabled using AI auto- segmented masticatory muscles, and the patient- centered importance in mitigating masticatory dose and reducing trismus. Keywords: trismus, auto-segmentation, dose/volume reduction
Made with FlippingBook - Share PDF online