A Novel Thyroid Nodule Recurrence Prediction Using Machine Learning Techniques

Abstract

One of the most prevalent endocrine diseases is represented by thyroid nodules, and most of them are benign. Nevertheless, there exists a large sub-population that has a great potential of recurring even after the initial treatment especially those related to the papillary thyroid carcinoma (PTC). It is essential to identify patients who are at increased risk of recurrence so that to intervene medically in time and manage patients individually. The conventional diagnostic procedures and risk stratification models frequently fail to detect sophisticated patterns in a multi-dimensional medical data.
This paper introduces a novel machine learning strategy of predicting whether thyroid nodules will come back or not using structured data as clinical and pathological information. Several supervised learning algorithms were tested, such as Random Forest, Gradient Boosting, Support Vector Machine, or LightGBM. Of them, the LightGBM model achieved the best results with the accuracy of 97.40 percent and the ROC-AUC of
99.17 percent. Age, gender, smoking history, thyroid functions, focal vs. diffuse nature of the tumor, tumor stage and response to initial treatment are some of the features that were used during model training.
The usage of data-based methods in the prediction of thyroid cancer makes this study clinically applicable to the develop- ment of personal treatment plans and long-term post-therapy monitoring. The given model can catalyze the performance of endocrinologists and oncologists in terms of decision-making process, which will lead to patient outcomes.