Prediction of Benign and Malignant Thyroid Nodules Based on Machine Learning
Jiaoxia Zhang, Sihui Chen · 2024
Thyroid nodules are a common endocrine organ disease, but up to 30% of thyroid nodules cannot be accurately classified as benign or malignant through cell pathology. In order to reduce the number of unnecessary surgeries for patients with benign nodules, this article predicts the benign and malignant nature of thyroid nodules. Firstly, fuzzy matching was performed on 2650 sample data from 1143 patients, and missing values were processed for 87 protein matrix indicators related to thyroid nodules. Secondly, a feature extraction technique based on genetic algorithm was adopted to select the indicators that are more likely to display malignant thyroid nodules from the preliminary extracted indicators, and random forest model, SVM model, XGBoost model, logistic model, and logistic XGBoost model were established. Finally, through comparison, the logistic based XGBoost model has a recognition accuracy of 80.2% and an AUC value of 87.2%, indicating the best classification performance.