Fig. 2: Thyroid gland and nodule segmentation with deep learning networks.
From: A fully autonomous robotic ultrasound system for thyroid scanning
a Thyroid and nodule segmentation network based on pre-training and weakly supervision. The feature and spatial losses are used to provide prior knowledge to the network considering the diversity of nodule samples and the space constrain between thyroid lobes. b Thyroid lobe segmentation based on pre-training. c Our proposed VariaNet and its variants predict results for different types of nodules. d Comparisons with the existing segmentation models on the SCUTN1K testset (best result in bold). e Receiver Operating Characteristic (ROC) curve of different algorithms on the SCUTN1K testset. Source data are provided as a Source Data file.
