INTEGRATING MACHINE LEARNING IN ENT FOR PREDICTIVE MODELING OF THYROID CANCER RISK AND RECURRENCE

Abstract: Background: Thyroid cancer recurrence presents a major clinical problem in managing patients due to the lack of reliable means of identifying patients who may be at high risk for recurrence to offer them appropriate early intervention. Traditional methodology often involves manual interpretation and a fusion of different types of data is not incorporated. In this work, the applicability of machine learning for thyroid cancer risk and its recurrence in the field of ENT by assessing clinical, genetic, and imaging data is discussed. Objective: The main purpose of this work is to perform machine learning to model thyroid cancer relapse and to determine significant predictors from multiple data sources to improve patients’ prognosis and tailor treatment plans. Methods: A retrospective observational study approach was implemented using EHRs, genetic data, and imaging studies (ultrasound, CT, etc.), in patients diagnosed with thyroid cancer. Among the Machine learning algorithms applied include Logistic Regression, Random Forest, Support Vector Machine (SVM), and Neural Networks were used to create predictive models. We used a random sample split into training and testing groups and calculated accuracy, sensitivity, specificity, and AUC-ROC. Chi-square and ANOVA statistical analyses and multivariate logistic regression were used to evaluate the relationship of risk factors to recurrence. Results: As for the indexes of accuracy, the Neural Networks and Random Forest models seemed more accurate, 0.88 and 0.87, accordingly. Five statistically significant markers of recurrence were identified they are tumor size odds ratio = 2.45, age odds ratio = 1.12, genetic mutations odds ratio = 1.67. The ROC analysis suggested very high model performance in all studies with AUC values close to 1. When it came to the risk groups, there were considerable differences regarding recurrence rates (p=0.002). Conclusion: H formula showed that Neural Networks and Random Forest two types of machine learning models present a 95 percent accurate method of predicting thyroid cancer recurrence. These models can therefore improve clinical decision-making in that first the best model is selected and used to assess high-risk patients to formulate a treatment plan. Future studies should extend to examine these models’ performance using actual-time data to refine the application of the models in aiding the management of thyroid cancer.