Application of artificial intelligence in ultrasound diagnostics of thyroid nodules

Abstract

BACKGROUND: the use of artificial intelligence in ultrasound diagnosis of thyroid nodules is expected and quite promising. However, in order to understand this, it is necessary to see how a doctor works with its help, diagnosing diseases step by step, and how exactly this intelligence is implemented in practical healthcare. The current publication provides an overview of existing intelligent systems for supporting medical decisions in thyroidology, and describes in detail the capabilities of the Russian intelligent computer assistant for ultrasound diagnostics - a system for stratifying thyroid nodules by EU-TIRADS categories.AIM: increasing the accuracy and reducing the time of ultrasound diagnostics in the study of thyroid nodules through the use of an intelligent system for assisting the ultrasound doctor at various stages of his activity with demonstration of the actions of the “assistant”.MATERIALS AND METHODS: to understand the possibilities of ultrasound doctors using artificial intelligence in their work, the proposed solution is divided into stages, each of which demonstrates the additional capabilities that a doctor has when using intelligent computer vision methods. Various artificial neural network architectures are used as an intellectual base, which can be further trained like a human on new medical data.RESULTS: the proposed intelligent solution allows the ultrasound doctor to have a “second opinion” at his workplace, which, by processing ultrasound cine loops, allows him to solve the problems of segmentation and stratification of thyroid nodules according to EU-TIRADS categories with an accuracy of 70%, i.e. at the level of a doctor with 5 years of experience. The proposed data driven approach will improve its accuracy as new patient loops are processed.CONCLUSION: the narrative leads the reader to understand in what diagnostic processes it is useful to use artificial intelligence methods in the ultrasound diagnosis of thyroid nodules, and how natural and artificial intelligence can effectively interact within the framework of a software web application.