A recent study presented a novel approach to diagnosing multiple sclerosis by integrating optical coherence tomography with images from infrared reflectance scanning laser ophthalmoscopy.
Optical coherence tomography (OCT) parameters have become biomarkers for multiple sclerosis (MS) progression, according to research published in Translational Vision Science & Technology. The peripapillary retinal nerve fiber layer (RNFL) and the ganglionic cell and inner plexiform layers in the macular region are thinner in MS patients compared to healthy controls. These reductions in thickness are associated with visual problems, different MS subtypes, physical and cognitive disabilities, and MRI findings in patients. Not only can these biomarkers be used to quantify neurodegeneration in MS, but they also help monitor progression and assess therapy efficacy.
Infrared scanning laser ophthalmoscopy (IR-SLO) is also used for retinal imaging. It creates two-dimensional images of the retina using a low-powered laser. "IR-SLO is usually performed along with OCT B-scan acquisition; this approach allows for accurate alignment of the B-scans despite eye movements, which improves the signal-to-noise ratio and reduces measurement variability at follow-up examinations,” noted study investigators.
They trained convolutional neural networks with IR-SLO images and OCT thickness maps as distinct input datasets. The top-performing models for each type were then integrated to develop a bimodal model.
They then analyzed a dataset of IR-SLO images and OCT data from 32 MS patients and 70 healthy controls using the neural networks to enhance diagnostic performance. The machine learning model with combined OCT and IR-SLO data had an accuracy of 96.85% (± 0.45%) in detecting MS. The model also showed 100% sensitivity and 94.96% specificity (± 0.66%), with a 99.69% area under the receiver operating characteristic curve (± 0.12%).
Integrating OCT and IR-SLO images significantly enhanced the diagnostic accuracy for MS iby leveraging the complementary information provided by both imaging modalities, compared to using OCT alone, and improved the model's ability to distinguish between MS and healthy states (approximately 3% higher than OCT thickness maps only).
"Notably, the model trained solely with OCT thickness maps outperformed the model that relied only on IR-SLO images, suggesting that IR-SLO images may lack sufficient features for distinguishing between MS and healthy subjects," noted investigators.
The researchers also examined how bimodal systems can assist in detecting optic neuritis, a key indicator of MS. They pointed out that nearly all MS patients have optic nerve neuropathy, whether or not they exhibit clinical symptoms, as confirmed by postmortem studies. During an eye exam, clinicians can only detect RNFL damage from optic neuritis when more than 50% of the layer is affected. This highlights the potential of AI-based systems to identify subtle optic nerve pathologies that might not be easily visible to humans in IR-SLO images.
Further studies with larger and more diverse datasets are needed to confirm these findings and enhance their applicability in real-world settings.
A full list of author disclosures can be found in the published research.
