AI Eye Scans a New Diagnostic Tool for MS?

A machine learning model trained with optical coherence tomography (OCT) and infrared scanning laser ophthalmoscopy (IR-SLO) retinal images can detect multiple sclerosis (MS) with “astonishing” accuracy near 100%, potentially leading to earlier diagnosis and treatment, researchers reported.

“This approach fills an unmet need in MS diagnosis by leveraging high-resolution en face images (IR-SLO) alongside OCT data,” study investigator Rahele Kafieh, Department of Engineering, Durham University, Durham, England, told Medscape Medical News.

“The improved diagnostic performance, with high sensitivity and specificity, suggests that this method can better differentiate between MS and healthy individuals, addressing the need for more accurate and reliable diagnostic tools in MS,” Kafieh added.

The study is published in the July issue of Translational Vision Science & Technology.

Two Scans Better Than One

Damage from MS can affect the retina. OCT can help quantify neurodegeneration in MS, monitor disability progression, and assess the efficacy of neuroprotective therapies. However, it remains unclear how incorporating IR-SLO, also known as monochromatic fundus imaging, may the enhance automated diagnosis of MS.

To investigate, researchers trained computer models to classify MS using IR-SLO and OCT data from 32 patients with MS and 70 healthy individuals.

In an internal test dataset, a biomodal model incorporating both IR-SLO and OCT images had an accuracy of 92%, a sensitivity of 95%, a specificity of 92%, and an area under the receiver operating characteristic curve (AUROC) of 97%.

The bimodal model held up in an external dataset, with an accuracy of 85%, a sensitivity of 97%, a specificity of 85%, and an AUROC of 99%.

“Integration of IR-SLO images and OCT thickness maps led to superior model performance (approximately 3% higher than when only OCT thickness maps were utilized),” the study team reported.

“This is indeed in line with our expectations, as the merged model leverages a broader range of input images from two distinct modalities, thus incorporating more useful information for accurate detection of MS,” they said.

“While the results are promising, this approach is not yet ready for clinical use,” Kafieh said.

To translate this into clinical practice, further studies with larger and more diverse populations are needed to validate the findings and ensure the model’s robustness and generalizability, she explained.

There is also a need to develop user-friendly software that integrates with existing clinical imaging systems to facilitate the use of this bimodal approach in routine practice and obtain regulatory approvals.

“By addressing these steps, the proposed bimodal model can be effectively translated into clinical use, offering a significant advancement in the diagnosis and management of MS,” Kafieh said.

Need for Further Research

Commenting on the research for Medscape Medical News, Patricia K. Coyle, MD, professor of neurology, Renaissance School of Medicine at Stony Brook, in Stony Brook, New York, who wasn’t involved in the study, echoed the need for further research.

“People are looking for a proven MS diagnostic biomarker, but we do not have that yet,” Coyle said.

This new research is “preliminary and would require extensive additional validation studies before it could be accepted. The numbers are very small, and there is not a robust control group of other neurologic disease subjects,” Coyle cautioned.

Also offering perspective on this research, Daniel Ontaneda, MD, PhD, staff neurologist with the Cleveland Clinic Mellen Center for Multiple Sclerosis, Cleveland, said, “It’s very helpful to know if there is optic nerve involvement as it provides greater confidence in the diagnosis of MS.”

“Conceptually, it’s a very attractive idea of combining OCT and infrared reflectance scanning laser opthalmoscopy, and this study is a great first step, but I don’t think it’s a definitive study,” Ontaneda told Medscape Medical News.

He noted that while most MS centers have OCT, infrared reflectance laser ophthalmoscopy is not as commonly available, which might be a barrier.

“It’s commonplace for us to do OCT in individuals who are presenting for a diagnosis of MS. OCT is not currently part of the diagnostic criteria but will likely be included in revised diagnostic criteria,” Ontaneda said.

Kafieh and Coyle had no relevant conflicts of interest. Ontaneda is the principal investigator in a diagnostic biomarker study in MS sponsored by the National Institute of Neurological Disorders and Stroke.