A Deep-Learning Based Ultrasound Text Classifier for Predicting Benign and Malignant Thyroid Nodules

The diagnosis of benign and malignant thyroid nodules timely and correctly is always a core problem for the clinical practice of thyroid nodules. Traditionally, preoperative diagnosis process of benign and malignant thyroid includes two stages: ultrasound check and fine needle aspiration. The malignant thyroid nodules will be further confirmed by surgery and pathology. However, such traditional diagnosis process falls into trouble in clinical practice since fine needle aspiration has potential risk of turning benign nodules into malignant ones. Moreover, the correct diagnosis of malignant thyroid nodules is to a great extent determined by the expertise of clinicians. As a new machine learning method, deep learning has been applied in computer-aid diagnosis recently. Thus, in this paper, we propose a deep-learning based ultrasound text classifier for predicting benign and malignant thyroid nodules. The proposed ultrasound text classifier is a kind of supervised classification based on deep neural network which is trained by the labeled ultrasonic text with benign or malignant label of pathology. Experimental results show that this method has the highest accuracy rate of 93% and 95% both on the real medical dataset and the UCI standard dataset, compared with the traditional Random Forest, Support Vector Machine and Neural Network both on the real medical dataset and the UCI standard dataset.