Semi supervised segmentation of thyroid based on ultrasound images with wavelet and boundaries features

This paper describes a method of semi-supervised segmentation of thyroid based on ultrasound images with wavelet and boundaries features. Discrete Wavelet Transform and Inverse Discrete Wavelet Transform were used to extract image frequency features and semantic information, while Convexity Loss Function and Active Contours Loss Function were used to evaluate boundaries features, which can improve the segmentation accuracy. 3168 thyroid ultrasound images have been segmented using the proposed method. DICE is 93.39%, IOU is 87.69%, PA is 96.65% and Convexity is 91.17%. It can be seen from contrast experiments that the method proposed in this paper can improve DICE by 5.56%, IOU by 8.95%, PA by 4.66% and Convexity by 2.43%.