AUTHOR=Cao Chun-Li , Li Qiao-Li , Tong Jin , Shi Li-Nan , Li Wen-Xiao , Xu Ya , Cheng Jing , Du Ting-Ting , Li Jun , Cui Xin-Wu 

TITLE=Artificial intelligence in thyroid ultrasound

JOURNAL=Frontiers in Oncology

VOLUME=Volume 13 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2023.1060702

DOI=10.3389/fonc.2023.1060702

ISSN=2234-943X

ABSTRACT=Objective: Artificial intelligence (AI) has made remarkable progress in thyroid ultrasound image recognition. The research on AI recognition of thyroid ultrasound images has been widely carried out. The purpose of this article is to classify and review the application of AI in thyroid ultrasound images.
Materials and Methods: The paper introduces the basic theoretical knowledge of AI, including traditional machine learning (ML) algorithms and deep learning (DL) algorithms, and their clinical application of ultrasonic imaging in thyroid diseases, such as thyroid disease detection and differential diagnosis of thyroid nodules, thyroid segmentation, etc. Finally, we gave comments on the challenges and prospects of AI in the clinical application of thyroid ultrasound.
Results: After review and analysis, the dominant AI algorithm is DL, especially the DL algorithm based on the convolutional neural network (CNN). However, the DL algorithm needs a number of data sets and a long training time. The commercial S-detect technology is the first computer-aided diagnosis (CAD) system for thyroid ultrasound examination using DL. It is more suitable for inexperienced operators or clinicians, but has no apparent clinical advantages for professional clinicians. In addition, the ML algorithm also plays a critical part in the ultrasonic diagnosis of thyroid diseases, and the support vector machine (SVM) is widely used.
Conclusion: As an advanced technology, AI can automatically and quantitatively evaluate image information, increase the accuracy and efficiency of image diagnosis, and improve the variability and subjectivity of traditional thyroid ultrasound diagnosis. To use AI wisely, radiologists should know the advantages and limitations of different algorithms and make up for their shortcomings through further research. We believe that in the foreseeable future, AI will facilitate radiologists to make better clinical decisions and become the primary development trend of ultrasonic diagnosis of thyroid diseases in the future.