Personalized Prediction of Lymph Node Involvement in Head and Neck Squamous Cell Carcinomas Using Mixture Hidden Markov Models Incorporating Tumor

Head and neck squamous cell carcinomas (HNSCC) frequently metastasize to regional lymph nodes, making accurate prediction of lymphatic spread crucial for treatment planning. Current diagnostic imaging techniques fall short in detecting microscopic lymph node metastases, often leading to broad and non-personalized irradiation strategies. In this study, we present an advanced model for predicting the risk of occult nodal disease by integrating the primary tumor location into a hidden Markov model (HMM) framework. We focus on tumors in the oropharynx and oral cavity, incorporating detailed subsites as defined by ICD-10 codes. By leveraging multi-centric data from over 1,200 patients, we developed a mixture of HMMs that account for the distinct patterns of lymphatic spread observed in different tumor subsites. Our approach enhances the precision of lymph node involvement predictions, potentially allowing for more personalized and targeted radiation therapy. The results indicate that our mixture model outperforms traditional models, especially in cases where lymph node involvement patterns vary significantly across subsites. This work represents a significant step towards personalized treatment planning in HNSCC, with the potential to reduce treatment-related side effects while maintaining therapeutic efficacy.