Segmentation of Ultrasound Thyroid Nodules Based on Encoder-Decoder Structure

This paper compares four ultrasonic segmentation methods of thyroid nodules. This research uses the encoder-decoder neural network structure to learn deeper semantic information in the down-sampling stage, gradually restore the position information in the up-sampling stage, and finally use skip connection integration features to complete the semantic segmentation of thyroid nodules. This paper tried four encoder-decoder structures on the ultrasound database of thyroid nodules. These experiments show that the structure is effective (maximum mIOU is 81.2
^%
).