Neural networks to identify multiple sclerosis with optical coherence tomography

Area of Science:

• Ophthalmology
• Neuroscience
• Artificial Intelligence

Background:

• Multiple sclerosis (MS) is a demyelinating disease affecting the central nervous system.
• Retinal nerve fibre layer (RNFL) thinning is a recognized biomarker for neurodegeneration in MS.
• Early and accurate detection of RNFL damage is crucial for managing MS progression.

Purpose of the Study:

• To compare axonal loss in ganglion cells using spectral-domain optical coherence tomography (OCT) in MS patients versus healthy controls.
• To evaluate the capability of an artificial neural network (ANN) technique to enhance the detection of RNFL damage in MS.
• To assess the diagnostic performance of ANN-based RNFL analysis.

Main Methods:

• A cohort of 106 MS patients and 115 healthy controls underwent OCT imaging.
• Circumpapillary RNFL thickness was measured using the Spectralis OCT system.
• An ANN was trained and evaluated for its ability to discriminate between MS and healthy eyes, with performance compared to standard OCT parameters using ROC curves.

Main Results:

• The ANN demonstrated a strong capability in detecting RNFL loss in MS patients, achieving an area under the ROC curve (AUC) of 0.945.
• The ANN outperformed individual OCT-derived parameters (mean and sector thicknesses) in discriminating between MS and healthy eyes.
• RNFL thickness measurements from OCT showed good differentiation ability between MS and healthy individuals.

Conclusions:

• Spectralis OCT measurements of RNFL thickness are effective in differentiating individuals with MS from healthy controls.
• The ANN technique significantly improved the detection of RNFL damage in MS patients compared to traditional OCT parameters.
• ANN-based analysis holds promise for enhancing the early diagnosis and monitoring of neurodegenerative changes in MS.