The Research of Automatic Classification of Ultrasound Thyroid Nodules

Abstract

This paper proposes a computer-aided diagnosis system which can automatically detect thyroid nodules (TNs) and discriminate them as benign or malignant. The system firstly uses variational level set active contour with gradients and phase information to complete automatic extraction of the boundaries of thyroid nodules images. Then according to thyroid ultrasound images and clinical diagnostic criteria, a new feature extraction method based on the fusion of shape, gray and texture is explored. Due to the imbalance of thyroid sample classes, this paper introduces a weight factor to improve support vector machine, offering different classes of samples with different weights. Finally, thyroid nodules are classified and discriminated by the improved support vector machine. Experiments show that the efficiency of discrimination on benign and malignant thyroid nodules is improved.