AI model analyzes images of eye to help accurately diagnose NMOSD
It outperformed or matched specialists depending on the scenario
A new artificial intelligence (AI) model that analyzes images of the eye can help accurately identify neuromyelitis optica spectrum disorder (NMOSD), a study shows.
The model outperformed specialists when relying solely on eye images. When clinical data such as MRI scans and antibody tests were provided, the model’s accuracy matched that of experienced experts — while continuing to outperform more junior doctors and residents.
“Our study demonstrated that the proposed [AI model] exhibits high sensitivity and specificity in identifying NMOSD patients, while also showing potential for predicting disease onset and progression,” researchers wrote.
The tool could help bring earlier and more accurate diagnosis to people in underserved areas.
“This innovation has the potential to assist governments in providing more accurate and timely medical support to economically disadvantaged populations, improving healthcare accessibility and outcomes,” researchers wrote.
The study, “Multimodal AI diagnostic system for neuromyelitis optica based on ultrawide-field fundus photography,” was published in Frontiers in Medicine.
Use of AI in NMOSD diagnosis remains unexplored
NMOSD occurs when the body’s immune system mistakenly attacks the spinal cord and optic nerve, which sends and receives signals from the eyes, resulting in symptoms like vision loss, loss of sensation, pain in the spine or limbs, and muscle spasticity. In most cases, these immune attacks are driven by antibodies known as AQP4-IgG, which target support cells in the nervous system called astrocytes.
“If left untreated, [NMOSD] can lead to severe and irreversible visual impairment as well as significant motor dysfunction,” the researchers wrote. “Hence, advancements in early and accurate diagnosis are crucial, as they would facilitate prompt therapeutic intervention.”
NMOSD diagnosis relies on a complex combination of specialist clinical evaluation, MRI scans, and blood tests for antibodies such as AQP4-IgG. These resources, however, may be limited in many parts of the world, particularly in rural or low-income areas. Additionally, some patients do not test positive for AQP4-IgG antibodies, which can further complicate and delay diagnosis.
In recent years, ultrawide-field fundus (UWF) photography has become a noninvasive tool for capturing detailed images of the back of the eye. At the same time, AI “have gained prominence in medical applications, particularly for [eye] disease detection,” the researchers wrote.
“However, despite NMOSD being a major cause of visual impairment worldwide, the application of AI in its diagnosis remains relatively unexplored,” the researchers added. “This raises an intriguing question: Could NMOSD be accurately diagnosed solely through UWF imaging, without the need for multiple diagnostic modalities?”
AL model tested against 6 ophthalmologists
In this study, a research team in China developed an AI model trained to recognize NMOSD-related features using 1,618 eye images. These included 330 images from 285 people with NMOSD and 1,288 images from 770 individuals without the condition. The data were collected between January 2022 and April 2024.
When tested using only the images, without any clinical data, AI achieved high diagnostic accuracy, reaching more than 90% of sensitivity, meaning it correctly identified most patients with NMOSD. It also reached more than 92% specificity, meaning it rarely misclassified healthy individuals as having the disease.
When clinical data were added to the model alongside eye images, its diagnostic accuracy improved further, reaching 97% sensitivity and 96.9% specificity.
To compare AI’s ability with those of human experts, the researchers tested the model against six ophthalmologists, or eye doctors: two senior specialists, two attending physicians, and two residents.
With improved computing capabilities and advanced database technologies, this model could potentially be developed into a comprehensive virtual screening system for NMOSD in clinical practice.
In a test using 200 images without any clinical context, the AI achieved an accuracy of 90.2%. In comparison, senior specialists scored 78.3%, attending physicians 56.8%, and residents 33.3%.
When clinical information was provided, the AI and senior specialists performed nearly equally (97.8% vs. 98%), and better than attending physicians (85.7%) and residents (70.2%).
“Our [AI model], designed to diagnose NMOSD using UWF images, streamline[s] the diagnostic process without requiring direct involvement from ophthalmologists or neurologists,” the researchers wrote. “With improved computing capabilities and advanced database technologies, this model could potentially be developed into a comprehensive virtual screening system for NMOSD in clinical practice.”
About the Author
