The goal of this observational study is to prospectively validate the efficacy of an AI multimodal model constructed based on multi - sequence orbital MRI in the diagnosis, activity and severity assessment, and prognosis prediction of thyroid - associated ophthalmopathy (TAO) in real - world clinical scenarios. The main questions it aims to answer are:

Can AI models accurately assess the presence, activity, and severity of Thyroid-Associated Orbitopathy (TAO)? Can AI models predict the prognosis of TAO? Researchers will compare the diagnostic accuracy of the AI model for TAO patients and healthy subjects to evaluate its diagnostic performance.

Participants will undergo a standardized, study - specific multimodal orbital MRI scan (sequences include T1WI, T2WI, STIR, and research sequences such as Magic, IDEAL - IQ, DWI, ASL, CEST). And will systematically acquire ocular ultrasound images from TED patients (active and inactive stages), non-TED ophthalmic disease controls, and healthy volunteers. AI-driven deep learning techniques (convolutional neural networks) will be applied to achieve automatic segmentation of key structures (extraocular muscles, optic nerve, lacrimal gland, and retrobulbar soft tissue). High-throughput radiomic features encompassing morphological parameters and gray-level texture patterns will be extracted. Machine learning algorithms will then be employed to construct objective prediction models for TED screening and activity staging, with MRI findings and CAS scores serving as the reference standards for external validation.