Early Assessment of Multiple Sclerosis by Spectralis Optical Coherence Tomography and Applying Convolutional Neural Network

Abstract

Introduction: Multiple Sclerosis (MS) is a prevalent autoimmune and inflammatory disorder that leaves demyelination and neurodegenerative changes in Central Nervous System (CNS). The retina is among body organs that is affected by MS, particularly the pRNFL, which is impaired during the early episodes of the disorder. Optical Coherence Tomography (OCT) images can play a key role in the preliminary stages. Convolutional Neural Networks (CNN)-based methods are commonly applied in image classification and have shown promising and applicable results in MS diagnosis. Method: in total, 197 MS patients and 283 healthy cases were included in this study, and Spectralis OCT images were taken then, using data augmentation, the CNN was trained with 15,000 images. Finally, the automatic diagnosis algorithm for MS disease was implemented in Python, and then the network loss processes diagram was drawn, and the sensitivity, specificity, and accuracy of the algorithm were evaluated. Result: The disease was successfully diagnosed by OCT images with an accuracy of 93.0, a sensitivity of 96.47, and a specificity of 90.44. Conclusion: The proposed method showed improvements in early-stage MS diagnosis and with the potentiality to be used in either the diagnosis or prediction of the progression of other diseases that affect the CNS (e.g. Alzheimer's disease, bipolar disorder, etc.). Keywords: convolutional neural network, multiple sclerosis, optical coherence tomography, retinal layer thickness