Wavelet Features of the Thickness Map of Retinal Ganglion Cell-Inner Plexiform Layer Best Discriminate Prior Optic Neuritis in Patients With Multiple Sclerosis

Abstract

The goal of this study was to analyze wavelet features of topographic thickness maps of the retinal ganglion cell-inner plexiform layer (GCIPL) and evaluate their discrimination ability in patients with multiple sclerosis (MS) and a history of optic neuritis (ON). Twenty-nine patients with relapsing-remitting MS and a history of ON were recruited together with 63 age-and sex-matched controls (HC). There were 33 eyes with a history of ON (MSON), 25 non-ON fellow eyes (MSFE) and 63 HC eyes. Ultrahighresolution optical coherence tomography was used to image the macula of each participant, and the volumetric dataset was segmented to yield a topographic thickness map of the GCIPL. The thickness map was analyzed using discrete wavelet transform to extract useful features, which were further analyzed with three machine learning methods, including logistic regression (LR), logistic regression regularized with the elastic net penalty (LR-EN), and support vector machine (SVM), for classification between groups. LR-EN with nested leave-one-out cross-validation resulted in 94% (83%) sensitivity and 100% (97%) specificity in discriminating MSON (MSFE) eyes from HC eyes in the training data and 88% (64%) sensitivity and 94% (94%) specificity in discriminating MSON (MSFE) eyes from HC eyes in the validation data, which were similar to or better than those resulting from LR and SVM. LR-EN improved the sensitivity and specificity of discriminating prior ON and MSFE eyes in patients with MS from HC eyes compared with traditional thickness analysis. It may be promising in facilitating the diagnosis of subclinical ON in patients with MS.

SLO-MSNet: Discrimination of Multiple Sclerosis using Scanning Laser Ophthalmoscopy Images with Autoencoder-Based Feature Extraction

“…Various input data have been utilized thus far, with the most desirable results achieved using the parameters obtained from MRI (pooled ACC = 96%), OCT (pooled ACC = 93%), CSF/serum (pooled ACC = 93%), and even gait and breathing pattern (pooled ACC = 88%) investigations. The OCT-based studies have mainly applied conventional ML classifiers on the thickness values (9,10,12,13,15–17,19) or the extracted features from them (11,14,18), with only one study utilizing DL (14). López-Dorado et al employed Cohen’s d coefficient technique on the thickness maps of RNFL, GCL+ (equivalent to GCIPL), GCL++ (equivalent to GCIPL plus RNFL), the total retina, and the choroid from 48 MS patients and 48 HC individuals.…”

“…Artificial intelligence (AI), encompassing both machine learning (ML) and deep learning (DL), has emerged as a promising aid for diagnosing MS (6), with its impressive performance being shown in a recent meta-analysis (7). Data analyzed for the automated classification of MS primarily stem from MRI, serum, CSF, and OCT investigations (8); specifically, the OCT parameters have involved the macular and/or peripapillary thickness of RNFL, GCIPL, inner nuclear layer (INL), and the whole retina, alone or in combination (9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19), leading to high levels of ACC reaching up to 100% (14).…”

“…Despite these advances, there remains a paucity of studies that assess the ACC of ML algorithms for automated diagnosis of neurodegenerative diseases, and in particular MS, using SLO images. To date, a number of studies have used OCT thickness measurements of different retinal layers for classification of MS using ML models (9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19). This is because a great body of evidence suggests that RNFL and GCIPL layers become thinner in MS patients, even in the absence of any history of ON (3)(4)(5).…”

Comprehensive Evaluation of Artificial Intelligence Models for Diagnosis of Multiple Sclerosis Using Information from Retinal Layers Multicenter OCT Images

“…Instead of utilizing conventional statistical analysis, AI approaches can be used to predict the progression of MS-associated disability. Traditional methods for analyzing retinal layer thicknesses obtained through OCT entail extracting relevant features from the images and subsequently classifying them using specific techniques such as support vector machines (SVM) [27,28], linear discriminant function (LDF) [29], artificial neural network (ANN) [30][31][32], decision tree [32], logistic regression (LR) and logistic regression regularized with the elastic net penalty (LR-EN) [33], multiple linear regression (MLR), k-nearest neighbors (k-NN), Naïve Bayes (NB), ensemble classifier (EC), long short-term memory (LSTM) [34], and recurrent neural network [35].…”

“…Recent scientific investigations have explored the utilization of machine learning (ML) methodologies specifically focused on analyzing data pertaining to the thickness of the RNFL in each of the four quadrants delineating the peripapillary region (pRNFL), as well as GCL++ (between inner limiting membrane to INL), and the macular ganglion cell-inner plexiform layer (GCIPL). Furthermore, additional factors including age, sex, best-corrected visual acuity, and other relevant data were duly considered in the analysis [27][28][29][30][31][32][33][34]. In our prior research [36], we devised a machine learning approach to overcome limitations present in earlier automated methods for distinguishing between individuals with MS and HCs.…”