ESTRO 2025 - Abstract Book

S1068

Clinical – Head & neck

ESTRO 2025

Purpose/Objective: Artificial intelligence (AI)-based imaging analysis has applications for the diagnosis and surveillance of head and neck malignancies. Serum circulating tumor-associated DNA (ctDNA) is an emerging biomarker used for response assessment and risk stratification in human papilloma virus (HPV)-associated oropharynx cancer, but with variable correlation to clinicopathologic factors including clinical stage. The utility of automated imaging biomarkers as a diagnostic complement to ctDNA has not yet been explored. Material/Methods: A secondary analysis was performed on a cohort of patients who were treated definitively for HPV-associated oropharynx cancers enrolled on a prospective blood collection protocol (Clinicaltrials.gov identifier: NCT04965792). All patients had pre-treatment serum measurement of tumor-tissue modified viral (TTMV) HPV DNA (NavDX, Naveris). Clinical data was abstracted from electronic medical records. An AI algorithm was applied to pretreatment diagnostic or radiation planning CT scans to auto-segment primary tumor and lymph nodes and calculate total volumes, as well as the volume of cystic or necrotic fluid within lymph nodes. Descriptive statistics were generated to describe patient demographics and disease characteristics. Univariable and multivariable linear regressions were performed to determine the association between ctDNA level (fragments/mL) and auto segmented tumor and nodal volumes, AJCC 8 th edition T and N stage, smoking pack years, HPV subtype (16 vs other), and Charlson Comorbidity Index (CCI). Model fit was assessed using Bayesian information criterion (BIC) and adjusted R 2 . Results: Between 2020-2023, 173 patients with HPV-associated oropharynx cancers were enrolled in the study. On univariable regression, automated tumor volume (coeff=38.84, p<0.001), automated nodal volume (coeff=38.08, p<0.001), T stage, N stage, HPV subtype, and CCI were associated with ctDNA (p<0.05 for each). Cystic nodal volume was not associated with ctDNA. On multivariable analysis, automated tumor and nodal volumes were independently associated with ctDNA (coeff=35.92, p<0.001 and coeff=24.42, p=0.021, respectively), but T and N stage were not (p=0.264 and p=0.821, respectively) when volumetrics were included in the model. Including automated tumor and nodal volumes improved model fit compared to T and N stage alone (3507.06 vs 3517.74 BIC, and 0.203 vs 0.111 R 2 ). Conclusion: AI-automated tumor volumetrics derived via pretreatment imaging are independently associated with circulating HPV tumor DNA, controlling for clinical stage. The association is stronger than clinical staging, and improved predictive capacity of regression models. AI-automated volumetrics may provide a practical way to estimate ctDNA levels and risk stratify patients, and in future studies, may be used alongside ctDNA to predict outcomes with standard or de-escalated therapy.

Keywords: circulating DNA, Artificial Intelligence, HPV

References: 1. Ferrandino, R. M., et al (2023). Performance of Liquid Biopsy for Diagnosis and Surveillance of Human Papillomavirus–Associated Oropharyngeal Cancer. JAMA Otolaryngology–Head & Neck Surgery, 149(11), 971. 2. Rettig, E. M., et al (2022). Association of Pretreatment Circulating Tumor Tissue–Modified Viral HPV DNA With Clinicopathologic Factors in HPV-Positive Oropharyngeal Cancer. JAMA Otolaryngology–Head & Neck Surgery, 148(12), 1120.