Prediction of cervical lymph node metastasis in papillary thyroid carcinoma using a machine learning approach

Abstract

The incidence of papillary thyroid carcinoma (PTC) has been constantly increasing over the past three decades, establishing it as the most frequently diagnosed type of thyroid malignancy. While patients with PTC generally have a favorable outcome, the presence of lymph node metastases (LNM) may significantly impact their prognosis, leading to a higher likelihood of recurrence. The current pre-operative diagnosis of LNM primarily relies on cervical ultrasound examination, which is limited in sensitivity. As a result of low sensitivity, lymph node metastases remain undetected on the pre-operative staging and may later present as persistent or recurrent disease, necessitating further evaluation and potential reoperation. To address the challenges of LNM diagnoses, various models have been developed to predict LNM in PTC patients. Among prediction models, special attention has been drawn to machine learning models that can predict disease outcomes with improved accuracy and enable individualized selection of optimal treatment for each patient. Therefore, this mini-review primarily focuses on explaining the fundamental principles of ML models through an example of LNM prediction in PTC patients. Additionally, an overview is provided on the most commonly used ML models in medicine, discussing their performance in studies employing such approaches for LNM prediction. Finally, the main challenges that limit the implementation of these models in clinical practice have been examined, and crucial areas for improvement have been identified. Currently, ML models present a potentially useful tool for LNM prediction in PTC patients, but further research is necessary to fully leverage their capabilities and enable their implementation into decision support systems.