Artificial intelligence in the diagnosis of thyroid diseases: applications and challenges

Abstract

Thyroid diseases, a prevalent class of endocrine system disorders, require diagnostic accuracy, which is essential for effective patient treatment and management. In recent years, artificial intelligence (AI) technology has made significant advancements in the medical field, providing new opportunities for the early diagnosis and precise treatment of thyroid diseases. This review discusses the latest applications of AI in the diagnosis of thyroid diseases, with a particular focus on the use of machine learning and deep learning (DL) algorithms in image classification, segmentation, and object detection within thyroid ultrasound, computed tomography, magnetic resonance imaging, and single photon emission computed tomography. Through the integration of cross-modal studies, this article reveals the application of AI across various imaging modalities, highlighting its potential value in feature extraction and risk stratification. Furthermore, we conduct an in-depth analysis of key challenges faced by AI applications, such as data heterogeneity (the decline in model performance due to data differences across institutions and equipment) and insufficient interpretability (DL models often function as "black boxes," making it difficult to provide transparent decision-making rationale, which limits clinical trust and adoption). In summary, AI technology demonstrates notable advantages and developmental potential in the automated diagnosis of thyroid diseases; however, its clinical translation still requires addressing the aforementioned challenges. The resultant analysis demonstrates that AI holds promise in improving the diagnosis and treatment of thyroid diseases, offering new pathways for personalized medicine and better patient outcomes. Specifically, AI-driven tools can reduce diagnostic variability and errors in thyroid nodule assessment, enhance treatment precision with risk-stratified recommendations, and support more consistent, individualized clinical decisions.

Ähnliche Arbeiten