Predicting recurrence in stage III and IV head and neck cancer patients undergoing chemoradiotherapy using PET/CT radiomics : A multicenter study using statistical and machine learning methods. Undergraduate Thesis

ABSTRACT : What can the metabolic behavior of a tumor tell us about a patient having recurrence after treatment? 18F-FDG PET/CT image is widely used in oncology because it allows doctors to detect tumors by observing regions in the body with abnormal glucose consumption. This “abnormal consumption” can be quantified through the Standardized Uptake Value (SUV). By setting different SUV thresholds within a delineated tumor, the analysis can go further by characterizing intra-tumoral metabolism. Head and Neck cancer (HNC) is the seventh most common worldwide and has a recurrence rate of 50%, therefore predicting recurrence is crucial for early treatment decision-making. In this study, a multicenter dataset containing 1872 clinical, tumor-to-lymph-node distances and SUV-intensity-based features extracted from the pre-treatment PET/CT images of 232 stage III and IV HNC patients was analyzed to exploit its capacity of predicting Disease-free survival (DFS). To achieve this, a comparative analysis was conducted by exploring combinations of dimensionality reduction techniques that went from the most traditional of correlation threshold shrinkage to PCA and even creating a completely novel approach of SUV threshold vs. feature curve (TFC) descriptors like the slope and the maximum value which are highly clinically interpretable. Then, also different state-of-the-art feature selection and machine learning survival models were tested until finding the combination with the best performance. This was achieved by the random survival forest model over the TFC descriptors + PCA dataset selected by univariate cox proportional hazards method that obtained a c-index of 0.7, outperforming the prediction capacity of the most used statistical method in survival analysis, which is multivariate Cox proportional hazards.

• Autores: Salas Villa, Eliana
• Tipo de recurso: Trabajo de grado de pregrado
• Fecha de publicación: 2025
• Institución: Universidad de Antioquia
• Repositorio: Repositorio UdeA
• Idioma: eng
• OAI Identifier: oai:bibliotecadigital.udea.edu.co:10495/44975
• Acceso en línea: https://hdl.handle.net/10495/44975
• Palabra clave: Neoplasias de Cabeza y Cuello; Head and Neck Neoplasms; Detección Precoz del Cáncer; Early Detection of Cancer; Tomografía Computarizada por Tomografía de Emisión de Positrones; Positron Emission Tomography Computed Tomography; https://id.nlm.nih.gov/mesh/D006258; https://id.nlm.nih.gov/mesh/D055088; https://id.nlm.nih.gov/mesh/D000072078
• Rights: embargoedAccess
• License: https://creativecommons.org/licenses/by-nc-sa/4.0/