Neural networks to identify multiple sclerosis with optical coherence tomography

Abstract

Purpose: To compare axonal loss in ganglion cells detected with spectral-domain optical coherence tomography (OCT) in eyes of patients with multiple sclerosis (MS) versus healthy control subjects using an artificial neural network (ANN). To analyse the capability of the ANN technique to improve the detection of retinal nerve fibre layer (RNFL) damage in patients with multiple sclerosis.

Methods: Patients with multiple sclerosis (n = 106) and age-matched healthy subjects (n = 115) were enrolled. The Spectralis OCT system was used to obtain the circumpapillary RNFL thickness in both eyes. The 768 RNFL thickness measurements provided by the Spectralis OCT were performed to obtain thickness measurements from 24 uniformly divided locations around the peripapillary RNFL. The performance of the ANN technique for identifying RNFL loss in patients with multiple sclerosis was evaluated. Receiver-operating characteristic (ROC) curves were used to display the ability of the test to discriminate between MS and healthy eyes in our population. ROC curves obtained using ANN and parameters provided by OCT (mean and 6 sector thicknesses) were compared.

Results: The capability of the ANN technique to detect RNFL loss in patients with multiple sclerosis compared with healthy subjects was good. The area under the ROC curve was 0.945. Compared with the OCT-provided parameters, the ANN had the largest area under the ROC curve.

Conclusions: Measurements of RNFL thickness obtained with Spectralis OCT have a good ability to differentiate between healthy and individuals with multiple sclerosis. Based on the area under the ROC curve, the ANN performed better than any single OCT parameter.

Keywords: multiple sclerosis; neural networks; optical coherence tomography; retinal nerve fibre layer.