Mohammad Amouzadeh,1,*Maryam Jafari,2Faeze Pourshakuri,3
1. Student research committee, Anzali International Medical Campus, Guilan University of Medical Sciences, Guilan,Iran 2. Student research committee, Anzali International Medical Campus, Guilan University of Medical Sciences, Guilan,Iran 3. Student research committee, Anzali International Medical Campus, Guilan University of Medical Sciences, Guilan,Iran
Introduction: Thyroid cancer is a malignancy that can spread to other parts of the body, most commonly via lymphatic spread in papillary carcinoma and blood spread in follicular and anaplastic carcinoma. Machine learning techniques show promise in predicting thyroid cancer metastasis by analyzing patient data. This comprehensive study aims to investigate the prediction of distant metastasis in thyroid cancer through the application of various machine learning (ML) models. Distant metastasis significantly impacts patient prognosis, necessitating early identification of high-risk individuals to optimize treatment strategies and improve survival rates. The ability to predict metastasis accurately can lead to more personalized care, allowing clinicians to tailor interventions based on individual patient risk profiles.
Methods: This research was conducted by searching keywords, including machine learning and metastasis prediction, and thyroid cancer, through databases, including Scopus and PubMed. Upon examining the articles, a comprehensive conclusion was derived from the collective findings.
Results: The study showed that the dataset was usually from the National Institutes of Health (NIH) Surveillance, local, and End Results (SEER) database, encompassing demographic and clinicopathological characteristics of thyroid cancer patients from 2010 to 2015. Multiple machine learning algorithms, including Random Forest (RF), Support Vector Machine (SVM), Decision Trees (DT), and Extreme Gradient Boosting (XGBoost), were utilized to construct predictive models for distant metastasis, cervical lymph node metastasis (CLNM), lung metastasis (LM), and bone metastasis (BM) in patients with papillary thyroid carcinoma (PTC) and thyroid cancer (TC). Through univariate and multivariate analyses, the study identified independent risk factors such as age, gender, histological type, and lymph node involvement. Among the evaluated models, RF consistently demonstrated superior predictive performance, achieving an area under the receiver operating characteristic curve (AUC) of 0.948 for distant metastasis, 0.889 for CLNM, 0.99 for LM, and 0.917 for BM. The models' effectiveness was further validated using metrics like sensitivity, accuracy, and F1 score. The BM prediction performance showed a sensitivity of 0.929, an accuracy of 0.906, and an F1 score of 0.908.
Conclusion: The findings indicate that ML-based prediction models can significantly aid in clinical decision-making by accurately identifying patients at risk for various types of metastases. The RF model, in particular, provides a robust framework for predicting outcomes, thereby facilitating personalized treatment strategies and improving patient management in thyroid cancer cases. The integration of these predictive models into clinical practice has the potential to enhance early diagnosis and intervention, ultimately leading to better patient outcomes.
Keywords: Machine learning
prediction
thyroid cancer
metastasis