Artificial intelligence (AI) can help diagnose thyroid eye disease and assess its severity by analyzing CT scans, research shows.
The technology could automate screening and help ensure patients who have compressive optic neuropathy from severe disease receive timely care, Paul Zhou, MD, and colleagues reported at the 2023 annual meeting of the Association for Research in Vision and Ophthalmology.
Thyroid eye disease, a rare autoimmune condition that leads to enlargement of extraocular muscles, fat, and connective tissue, can cause various eye-related problems that may reduce quality of life and threaten vision.
To develop a method of screening for thyroid eye disease using AI, Zhou, a researcher with Mass Eye and Ear in Boston, and his co-authors conducted a study using CT scans from patients seen at the Mass Eye and Ear.
Their dataset included hundreds of images from a control group of 20 eyes without orbital pathology, 60 eyes with thyroid eye disease but no evidence of compressive optic neuropathy, and 64 eyes with severe disease including features of compressive optic neuropathy. The researchers trained the AI model using 628 images; 157 images were used for testing.
The model had an overall accuracy of about 94%, the researchers reported. Two images from the control group were misclassified as having mild thyroid eye disease, and six images from the mild disease group were misclassified as normal. AI correctly classified all images from patients with severe disease.
The study shows that neural network-based analytic AI models may be able to help clinicians detect thyroid eye disease and screen for disease severity “entirely based on CT scans,” the investigators wrote.
Though some groups have embraced the technology, others worry that algorithms might have unrecognized biases. For example, researchers should ensure that AI does not interpret images from certain patient groups less accurately than it does others, experts have said.
The researchers had no disclosures.
Association for Research in Vision and Ophthalmology 2023 Annual Meeting. Presented April 23, 2023.