Classification of Thyroid Standard Planes in Ultrasound Images based on Multi-feature Fusion

Abstract

The automatic identification of thyroid ultrasound standard planes is of great significance in improving the efficiency of thyroid ultrasound diagnosis and treatment. This paper proposes a method for the automatic classification of thyroid ultrasound standard planes based on multi-features fusion. According to the characteristics that the thyroid ultrasound image is not affected by illumination, two local features about histograms of oriented gradients (HOG) and gray level co-occurrence matrix (GLCM) are extracted, and then the thyroid ultrasound images are adopted by SVM classifier, KNN classifier and Bayes classifier respectively. In the experiment, a total of 2111 thyroid ultrasound standard planes are classified. The results show that the SVM classifier is the best, and the accuracy of the four planes is 97%, 98%, 80% and 70% respectively. The classification accuracy rate is 86.25%, and the proposed method can provide a basic method for the automatic classification of thyroid ultrasound standard planes.