Mark-Guided Segmentation of Ultrasonic Thyroid Nodules Using Deep Learning
Shujun Zhou, Hong Qi Wu, Jie Gong, Ting Le, Hao Wu, Qin Chen, Zenglin Xu · 2018
Segmentation of thyroid nodules in the ultrasound image is a chal lenging task not only because of the speckle noise in ultrasound images but also the heterogeneous appearance and blurry bound aries of thyroid nodules. In this paper, we apply U-Net, a fully convolutional neural network, to thyroid nodule segmentation, and further proposed an interactive segmentation method based on it and the guidance of annotation marks. Firstly, the four end-points of the major and minor axes of a nodule are determined manually. Then, four white spots are directly drawn at the four points on the image to guide the training and inference of the deep neural network. Our method is evaluated on a dataset composed of 900 ultrasound thyroid images. The experimental results indicate that our mark-guided segmentation method is able to delineate nodules accurately with little human intervention and achieve a remarkable improvement over its automatic counterpart.