Fontes, Elisa Kauark; 0000-0001-7375-379X; http://lattes.cnpq.br/7843394158094803
Resumen:
Oral mucositis (OM) is a common side effect of head and neck radiotherapy (RT), resulting from a complex interplay of multiple risk factors. Strong evidence identifies RT dose as a key contributor to OM development. A detailed dose-response analysis of individual organs at risk (OARs) may help establish dose constraints to improve patient outcomes. Additionally, machine learning (ML) offers a promising approach by integrating both dosimetric and non-dosimetric factors for a more comprehensive risk assessment. This study aimed to assess OM risk prediction using ML and investigate the impact of dose distribution on OM development, identifying potential OARs related to OM. In the first study, an ML performance was tested to predict MO risk using a cross-validation strategy based on two dataset versions: one with all features and another with feature selection. Comparative analysis showed no relevant results with the full dataset, while feature selection improved performance, with the K-Nearest Neighbors algorithm achieving 64% accuracy, 58% sensitivity, and 68% specificity. The second study involved a dosimetric analysis of 57 head and neck cancer patients. Potential OARs for OM were identified, and dose-volume histograms were generated for OM onset and the final RT session, comparing Dmean and Dmax with OM incidence and distribution. Significant dosimetric differences were observed across all OARs except the upper lip. A Dmean cutoff of 48.4 Gy for the oral tongue was identified (92% accuracy, 96% specificity, 78% sensitivity). Additionally, each incremental 1 Gy increase in dose to the OARs was associated with a 1% higher risk of OM. These findings highlight the need for standardized OAR delineation to optimize RT planning and reduce OM incidence.
