AI raises the bar for diagnosing severe TED

Elvira Manzano
23 May 2023
AI raises the bar for diagnosing severe TED

Artificial intelligence (AI) can help diagnose thyroid eye disease (TED) and assess for severity by analysing CT scans, reports an expert at ARVO 2023.

“The technology could automate screening and help ensure patients who have compressive optic neuropathy from severe disease receive timely care,” said Dr Paul Zhou from the Mass Eye and Ear, a specialty hospital in Boston, Massachusetts, US.

TED is a rare autoimmune disease caused by the activation of orbital fibroblasts by autoantibodies directed against thyroid receptors. The condition leads to enlargement of extraocular muscles, fat, and connective tissue and can result in eye-related problems that reduce quality of life and threaten vision.

AI for TED

Using AI, Zhou and his team developed a screening method for TED in patients seen at his hospital.

Data included images from 20 eyes without orbital pathology, 60 eyes with TED but no evidence of compressive optic neuropathy, and 64 eyes with severe disease and compressive optic neuropathy features. Of this, 157 images were used for testing.

Accuracy was at 94 percent, with two images misclassified as mild TED. Six images which were supposedly mild disease were misclassified as normal. Interestingly, AI accurately identified images from patients with severe disease.

“Neural network-based analytic AI models could help clinicians detect TED and screen for disease severity using CT scans,” said Zhou.

AI screens for DR, too

In November last year, the US FDA approved another AI system to help diagnose diabetic retinopathy (DR), which is the leading cause of blindness in people with diabetes. []

An AI algorithm detects diabetic retinopathy within seconds, making referrals to an eye care specialist feasible for further evaluation and treatment.

For primary care clinicians, AI screening for retinopathy presents an opportunity to emphasize how important it is to manage the disease and what the consequences can be. For patients, AI is another tool to get them more engaged their own care.

Although many experts have embraced the technology, others worry that the algorithms may have some unrecognized biases. Hence, more research on AI for medical use is warranted, said Zhou.

Editor's Recommendations