The differentiation between benign and malignant thyroid nodules is pivotal in clinical practice. However, current diagnostic systems face challenges, such as heterogeneity in ultrasound interpretation, the invasiveness of fine-needle aspiration cytology and indeterminate of results. Artificial intelligence (AI) has evolved from traditional machine learning to deep learning, enabling advances in image recognition, feature extraction, and multimodal data integration. Nevertheless, AI applications remain constrained by limitations, including the insufficiency of high-quality annotated data and inadequate cross-center generalizability. Future progress will focus on algorithm refinement, multimodal data fusion, and multicenter clinical validation to enhance model generalizability and diagnostic accuracy.

The differentiation between benign and malignant thyroid nodules is pivotal in clinical practice. However, current diagnostic systems face challenges, such as heterogeneity in ultrasound interpretation, the invasiveness of fine-needle aspiration cytology and indeterminate of results. Artificial intelligence (AI) has evolved from traditional machine learning to deep learning, enabling advances in image recognition, feature extraction, and multimodal data integration. Nevertheless, AI applications remain constrained by limitations, including the insufficiency of high-quality annotated data and inadequate cross-center generalizability. Future progress will focus on algorithm refinement, multimodal data fusion, and multicenter clinical validation to enhance model generalizability and diagnostic accuracy.