2389 - Integration of Serum Metabolomics, MRI Radiomics, and Clinical Parameters for Predicting Long-Term Post-Radiotherapy Trismus in Nasopharyngeal Carcinoma: A Tri-Modal Modeling Approach

Purpose/Objective(s): This study develops a predictive model for long-term (= 3 years) post-radiotherapy trismus in nasopharyngeal carcinoma (NPC) patients. The model integrates three key components: serum amino acid metabolomics, radiomics features of masticatory muscles from MR images, and clinical data (including patient characteristics and radiation dosimetry parameters). This tri-modal approach aims to provide personalized risk assessment for radiotherapy-induced mouth opening difficulty.

Materials/Methods: This study analyzed 30 nasopharyngeal carcinoma (NPC) patients who underwent radical radiotherapy at the First Affiliated Hospital of Soochow University between 2017 and 2028. Pre- and post-radiotherapy blood samples were subjected to LC-MS analysis for amino acid profiling. Radiomics features were extracted from T1-weighted images (T1WI) of the masticatory muscles (masseter, temporalis, medial and lateral pterygoid muscles). Clinical data and radiation dosimetry parameters were collected. The primary endpoint was defined as = grade 2 mouth opening difficulty (CTCAE v5.0) at 3-year follow-up. LASSO regression identified key predictors, and model performance was evaluated using 5-fold cross-validation, comparing the tri-modal model against single-modal approaches.

Results: Of the 30 patients, 19 (63.33%) developed grade =2 mouth opening difficulty. Five amino acids (alanine, asparagine, glutamic acid, proline, and citrulline) exhibited significant post-radiotherapy decreases (P < 0.05), associated with protein synthesis, neurotransmitter synthesis, and redox reactions. From 1130 radiomics features, LASSO regression identified four key features for calculating patient-specific radiomics scores. The integrated tri-modal prediction model, combining changes in amino acid levels, radiomics features, clinical parameters, and radiation dosimetry, demonstrated superior performance (AUC = 0.87) compared to single-modal approaches (radiomics: 0.75, metabolic: 0.68), with good calibration (Hosmer-Lemeshow p = 0.42).

Conclusion: Post-radiotherapy trismus is a common long-term complication in nasopharyngeal carcinoma (NPC) patients, significantly impacting treatment compliance and quality of life. This pioneering study integrates metabolic reprogramming, radiomics of masticatory muscles, and clinical parameters to predict long-term mouth opening difficulty, demonstrating the value of this multimodal approach. While the study reveals important associations between amino acid profiles and trismus, larger prospective studies incorporating multi-omics and artificial intelligence (AI) are needed to validate these findings and develop targeted nutritional interventions.