Deep learning for image segmentation: Ultrasound image segmentation of thyroid nodules based on U_Net

The purpose of this article is to investigate the value of deep learning algorithms in the application of ultrasound images of thyroid nodules. Using a dataset of 7288 ultrasound images of thyroid nodules provided by the MICCAI 2020 Challenge, based on the U_Net framework, incorporating a multiscale input mechanism and improving loss optimization function, through continuous training to find the optimal model, so that the computer can autonomously segment the thyroid nodules. The segmentation accuracy reaches 0.955, and the network has good segmentation performance.