Research And Implementation Of Semi-automatic Marking System For Thyroid Nodules

The thyroid is an important endocrine organ of the human body and is closely related to the human body’s nervous,digestive,and cardiovascular systems.In thyroid ultrasound imaging diagnosis,the main basis for diagnosis by doctors is the characteristics of thyroid nodules.Compared with traditional medical image analysis methods,deep learning can automatically extract features,avoid tedious manual extraction steps,and surpass manual operations in terms of diagnostic efficiency and stability.In the actual application of deep learning,the acquisition of high-quality labeled data sets is still an indispensable part.However,in the process of constructing high-quality thyroid data sets,there are still three problems:(1)Manual labeling is inefficient and has a high error rate.(2)The labeling quality of the data set is uneven.(3)The labeling accuracy provided by the labeling tool is insufficient.In view of the above three problems,the segmentation algorithm of thyroid nodules under ultrasound images was studied,and a semi-automatic labeling system for thyroid nodules was realized.The research content of this paper is as follows1.Increasing the labeling rate:This paper proposes an improved DeepLabV3+model based on the cascade framework to realize the segmentation algorithm of thyroid nodules under ultrasound images.The model proposed in this paper has made two improvements on the basis of the original DeepLabV3+.One is to use the EfficientNet network to replace the original ResNet network for model feature extraction.The second is to improve the accuracy of model segmentation by means of stepwise segmentation.The improved model achieves a Dsc coefficient of 0.86,which can be applied to the automatic segmentation of nodules under thyroid ultrasound images,reducing the workload of labelers and improving labeling efficiency.2.Improving the quality of labeling:In addition to supporting basic labeling functions,the system in this paper also provides a set of online communication and collaboration mechanisms for team collaboration and labeling to achieve labeling quality control during the labeling process.3.Improving labeling accuracy:The front-end labeling function of the system is selected for secondary development on the open source framework Cornerstone Tools,and a variety of labeling tools are introduced,including a brush tool that supports free drawing,to meet the accuracy requirements of medical image labeling.In order to meet the needs of the platform,the source code of the framework is appropriately modified to adapt to the platform requirements.efficiency of labeling.