BACKGROUND
Thyroid Eye Disease (TED) is an autoimmune condition that can affect patients with Graves’ disease. TED occurs in about one in every three patients with Graves’, with most cases being mild. In TED, there is inflammation of the tissues around the eyes (especially eye muscles and fat) causing the eye to be pushed forward (bulging), eye pain, tearing, swelling of the conjunctiva (the white part of the eye), double vision and, if severe, it can affect the optic nerve and cause loss of vision. Making an early diagnosis of TED is very important so it can be monitored and treated before the disease permanently damages the eye.

Artificial intelligence (AI) is a new technology that allows machines to think, learn and make decisions, like humans. For example, by providing information such as X-rays to the machine, it learns to recognize and diagnose certain conditions and can even tell how serious the condition maybe. Many fields in medicine are now utilizing AI to help with early diagnosis and identifying diseases. This study looks at how good AI is in recognizing and assessing the severity of TED and compared it to ophthalmologists specialized in TED (oculoplastic surgeons).

THE FULL ARTICLE TITLE
Lin LY et al. A deep learning model for screening computed tomography imaging for thyroid eye disease and compressive optic neuropathy. Ophthalmol Sci 2023;4(1):1000412; doi: 10.1016/j.xops.2023.100412. PMID: 38046559.

SUMMARY OF THE STUDY
The authors looked at patients who presented at the Massachusetts Eye and Ear Institute in Boston from 2011-2021. Patients who had CT scans of the orbits and eye exams done confirming TED were included in the study (total of 123 patients). Patient who had CT scan but did not have TED served as normal controls (31 patients). Based on their eye exam, patients were classified as normal, mild TED or severe TED (optic nerve damage). The average age of the patients was 71, and the majority were women (74%) and white (75%). The machine learned to recognize TED by first being taught (trained) on how CT images of patients with no disease, mild and severe TED look like. Once trained, when the machine looked at the CT scans of the orbits of the study patients, it correctly identified the diagnosis (normal, mild or severe TED) in almost 90%. Overall, 3 patients with severe TED were wrongly classified as mild, but none of the patients with severe TED were misclassified as normal. When eye specialist looked at the same CT scans, they correctly made a diagnosis in 70% of the cases.

WHAT ARE THE IMPLICATIONS OF THIS STUDY?
This study shows that when using this design of machine learning, AI can correctly diagnose TED, including its severe form, in almost 90% of cases and does it better than experts. By being able to correctly diagnose severe TED in CT scans, it can alert the non-specialized physician to make an urgent referral to an oculoplastic surgeon. Timely treatment of severe TED can prevent vision loss.

— Susana Ebner MD

Table of Contents | PDF File for Saving and Printing