Intelligent Diagnostic Methods for Thyroid Nodules

According to size, shape, echogenicity and other ultrasonographic features, physicians can determine whether thyroid nodules are malignant or not. However, the diagnosis results might vary due to differences in physicians' expertise and experience. This paper focuses on ultrasound diagnosis of benign and malignant thyroid nodules by introducing several classical machine learning methods. These machine learning methods can establish intelligent diagnosis models to determine the conditions of new cases based on the knowledge learned from the existing ultrasound data and the confirmed diagnosis results. This paper compares and analyzes the diagnosis performance between intelligent methods and physicians as well as the performance of different machine learning methods in terms of the diagnosis accuracy, sensitivity, specificity and other metrics experimentally. Except for the decision tree, the accuracies of all of the intelligent diagnostic methods are approximately 0.84, and the AUCs are greater than 0.87, which are comparable with physicians' decision accuracies. Different intelligent methods have different characteristics. For example, some learn fast, and some have better interpretability. The study shows that machine learning-based intelligent diagnosis methods can provide physicians with clinical decision support in the diagnosis of thyroid nodules.