ABSTRACT
Thyroid dysfunction is one of the most common endocrine diseases. About 4.7% of the United States population has undiagnosed thyroid disease (Garmendia Madariaga, Santos Palacios, Guillén-Grima, & Galofré, 2014). Thyroid nodules could be present in up to 67% of the population (Ezzat, Sarti, Cain, & Braunstein, 1994). Increased incidence of thyroid nodules has been attributed to increased use of imaging modalities and improvement in imaging technology (Singh, Singh, & Khanna, 2012). Technological improvements in ultrasound including elastography, 3D ultrasound (Liang et al., 2019) 274and quantitative ultrasound (Goundan et al., 2019) have been used to improve the diagnostic accuracy. Computer-aided diagnosis software that can automatically detect different features in the thyroid nodules and generate a report has been cleared by the FDA (Lu, Shi, Zhao, Song, & Li, 2019). Similarly, artificial intelligence (AI) algorithms have been used in the diagnosis and management of thyroid diseases. One of the first papers using AI in the diagnosis of thyroid disease was by Sharpe et al. in 1993. They used a multilayer perceptron trained by back-propagation and a learning vector quantization network to investigate the robustness of these models on noisy diagnostic data. In recent years, most of the AI research in thyroidology has been focused on the diagnosis and management of thyroid nodules. Initial approaches used texture analysis to classify thyroid ultrasound images. Later, AI techniques such as machine learning (ML) and an advanced form of ML, deep learning (DL) algorithms were used. In this chapter, we discuss the use of AI in thyroid imaging, cytopathological diagnosis of thyroid nodules and molecular markers. The use of wearable devices and generative adversarial networks in thyroidology is also discussed in this chapter.
