Abstract
Background/objectives
Study of retinal structure based on optical coherence tomography (OCT) data can facilitate early diagnosis of relapsing-remitting multiple sclerosis (RRMS). Although artificial intelligence can provide highly reliable diagnoses, the results obtained must be explainable.
Subjects/methods
The study included 79 recently diagnosed RRMS patients and 69 age matched healthy control subjects. Thickness (Avg) and inter-eye difference (Diff) features are obtained in 4 retinal layers using the posterior pole protocol. Each layer is divided into six analysis zones. The Support Vector Machine plus Recursive Feature Elimination with Leave-One-Out Cross Validation (SVM-RFE-LOOCV) approach is used to find the subset of features that reduces dimensionality and optimises the performance of the classifier.
Results
SVM-RFE-LOOCV was used to identify OCT features with greatest capacity for early diagnosis, determining the area of the papillomacular bundle to be the most influential. A correlation was observed between loss of layer thickness and increase in functional disability. There was also greater functional deterioration in patients with greater asymmetry between left and right eyes. The classifier based on the top-ranked features obtained sensitivity = 0.86 and specificity = 0.90.
Conclusions
There was consistency between the features identified as relevant by the SVM-RFE-LOOCV approach and the retinotopic distribution of the retinal nerve fibres and the optic nerve head. This simple method contributes to implementation of an assisted diagnosis system and its accuracy exceeds that achieved with magnetic resonance imaging of the central nervous system, the current gold standard. This paper provides novel insights into RRMS affectation of the neuroretina.
Subject terms: Optic nerve diseases, Physiology
Introduction
Definitive multiple sclerosis (MS) diagnosis is currently performed according to the modified McDonald criteria, [1] which are principally based on oligoclonal bands and magnetic resonance imaging biomarkers. However, these only offer moderate sensitivity and specificity [2] and the rate of misdiagnosis can be as high as 16% [3]. Moreover, these tests are invasive for patients and expensive for healthcare systems. For these reasons, there is a need to find new biomarkers, preferably non-invasive ones like the biomarker proposed in this paper. The retina constitutes the unmyelinated extension of the axons of the anterior visual pathway and forms part of the central nervous system. The complex comprising the three inner layers of the retina (retinal nerve fibre layer, RNFL; ganglion cell layer, GCL; and internal plexiform layer, IPL) is known as the internal retinal layer (IRL), and can be visualised and measured using optical coherence tomography (OCT).
Numerous studies indicate that IRL thicknesses provide a valuable biomarker in the early diagnosis, monitoring, and prognosis of incipient MS, and their findings suggest that these thicknesses offer even greater precision than the diagnostic methods (cerebrospinal fluid analysis and magnetic resonance imaging) employed in the current gold standard [4–6]. Recent papers also propose inter-eye difference or retinal asymmetry as a possible biomarker of MS. In Petzold et al. [7] (71,939 control subjects and 144 MS patients without optic neuritis), differences in MS were detected in the macular areas of the three layers of the IRL complex. Bijvank et al. [8] observed that the diagnostic accuracy of inter-eye differences when differentiating between control subjects and MS patients without optic neuritis (ON) was moderate (AUC: 0.66–0.73) after analysing the differences in two layers (GCL + IPL in the macular area and RNFL in the peripapillary area). Ortiz et al. [5] report the capacity to discriminate between control subjects and MS patients by analysing both thickness and inter-eye differences. Although they observe that the AUC is greater when using the thickness values, an optimal automatic classifier is obtained when considering both feature types. Patil et al. [9] found that inter-eye differences in the peripapillary RNFL and the GCL + IPL in the macular area of the eye is stable over time, even in patients without ON.
In this study, we used artificial intelligence (AI) to analyse the OCT variables with the greatest discriminant capacity in order to implement an assisted MS diagnosis system. An important concern in medicine is the lack of information about how AI algorithms obtain their results. The tools available in eXplainable Artificial Intelligence (XAI) attempt to address this shortcoming by implementing algorithms that identify the role that each input feature plays in the decisions made by automatic classifiers (see [10] for a review). Using Recursive Feature Elimination (RFE) [11], it is possible to select those key features that condition the response of the automatic classifier. The potential advantages of this method are that it makes it possible to rank the importance of the extracted features, choose the optimal feature subset, reduce the dimensionality and potential overfitting of the classifier, and interpret its predictions.
The aim of this paper is to explore, using XAI techniques, the layers and areas suffering greatest structural affectation (thickness and inter-eye difference) captured in OCT recordings taken from control subjects and recently diagnosed relapsing-remitting multiple sclerosis (RRMS) patients with no history of ON. Its secondary objective is to assess the assisted diagnosis system used to input the most relevant features and explain the results.
Methods
This study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Clinical Research Ethics Committee of Aragon (Zaragoza, Spain). Participants provided written informed consent after receiving a detailed explanation of the study. Data from these patients were previously analysed by Ortiz et al. [5].
Based on our preliminary studies in MS patients, we calculated the sample size needed to detect differences of at least 3 μm in OCT-measured thicknesses by applying a bilateral test with α 5% risk and β 10% risk (i.e., with a power of 90%). In order to obtain a sufficient sample of MS patients to allow in-depth study of the natural history of the disease, the non-exposed/exposed ratio was determined to be 0.5. From these data, it was concluded that at least 50 eyes would be needed in each group.
Definitive MS diagnosis was based on standard clinical and neuroimaging criteria [1]. To ensure a homogeneous population, only patients with the RRMS phenotype (diagnosed by a neurologist specialised in MS and by EGM) and no history of ON in either eye were included. The healthy controls had no history of ocular or neurological disease and presented no signs or symptoms of them; and they were recruited from subjects coming for preventive occupational health consultations (check-ups of hospital workers), family members of researchers and of patients. The recruitment was prospective from January of 2019 to December of 2022.
All participants received a comprehensive ophthalmological examination, including visual acuity using the ETDRS acuity charts, refractive error, intraocular pressure, slit lamp biomicroscopy, and study of the posterior segment of the eye by fundoscopy and OCT visualisation, to check that they did not meet the exclusion criteria.
The exclusion criteria were best-corrected visual acuity lower than 0.5 according to Snellen charts, refractive errors higher than 5 dioptres of spherical equivalent refraction or 3 dioptres of astigmatism, intraocular pressure higher than 20 mmHg, media opacifications (nuclear colour/opalescence, cortical or posterior subcapsular lens opacity lower than 2 according to the Lens Opacities Classification System III, performed by EGM, ophthalmology) [12], concomitant ocular disease (including glaucoma or retinal pathology), and other systemic conditions potentially affecting the visual system.
Two patients were excluded from the study because had an optic nerve papilla of one of the two eyes with atypical morphology which made the OCT data unreliable.
Figure 1A shows the block diagram for the implemented method. The retinal thickness maps were obtained with the Posterior Pole Retinal Thickness Map protocol using a Spectralis OCT device (Heidelberg Engineering Inc., Germany). The 25° × 30° explored area is represented as an 8 × 8 grid (64 cells) that provides overall retinal thickness. Each layer was segmented with the inbuilt software (HRA version 6.0.7.0). Scan quality was analysed according to the OSCAR-IB Consensus Criteria for Retinal Quality Assessment [13]. No manual correction was applied to the OCT output and only high-quality scans (25/40) were used for analysis.
Fig. 1. Diagram showing the steps of the research study performed.
A General block diagram for the implemented method. B Identification of the analysis zones. SVM support vector machine, RFE recursive feature elimination, LOOCV leave-one-out cross validation, ST supero-temporal, SN supero-nasal, IN infero-nasal, IT infero-temporal.
The structures analysed in this study were the three inner retinal layers in the macular area (mRNFL, mGCL, mIPL) and the complex comprising those three layers (mIRL) since they correspond to the neuroretina and are closely related to the central nervous system.
Both eyes of each subject were evaluated. In each subject, and for each layer analysed, both the average thickness value for the R and L eyes (Avg) and the average inter-eye difference value (Diff), defined as ABS (R–L) were obtained for the six regions shown in Fig. 1 B: Zone 1 (central papillomacular bundle: central PMB), Zone 2 (peripheral papillomacular bundle), Zone 3 (supero-nasal quadrant), Zone 4 (infero-nasal quadrant), Zone 5 (infero-temporal quadrant) and Zone 6 (supero-temporal quadrant).
The retinotopic distribution of the ganglion cell axons at retina level corresponds to these 6 delimited zones (Fig. 2): Zone 1 corresponds to the axons between the fovea and the optic nerve without deviation (central PMB); Zone 2 corresponds to the axons tracing a slight curve between the macula and the optic nerve (peripheral PMB); and Zones 3, 4, 5, and 6 correspond to the fibres in the rest of the retina that trace a semicircular path and converge in the optic nerve.
Fig. 2. Representation of the retinotopic distribution of the retinal nerve fibres in a healthy right eye.
The optic nerve is indicated by a white disc and the macula is indicated by dark colouring. The zones delimited in the study are marked on the image. Each zone corresponds to the semicircular path that the retinal nerve fibres trace as they fan out towards the periphery.
Under these conditions, the initial number of OCT features totals 48 (6 zones × 4 layers × 2 thicknesses/differences). SVM-RFE combines a support vector machine (SVM) classifier (with linear kernel) and the recursive feature elimination method.
Training a SVM consists of finding the best hyperplane that separates the two output classes (MS patients, control subjects); the input data points closest to the hyperplane are the support vectors and as these points are the most difficult to classify they are used in the training process to select the optimal hyperplane.
The efficacy of the training can be evaluated using a leave-one-out cross-validation (LOOCV) approach. The implemented SVM-RFE-LOOCV method comprises the following phases: 1) train a linear SVM using all the features under an LOOCV approach: this obtains 69 + 79 models. In each LOOCV interaction the weight of each feature in the SVM classifier is obtained, as is the efficacy of the classification using the subject data not used in the training; 2) calculate the ranking criteria of each input feature based on the average value of the SVM weights in the LOOCV; 3) discard the feature with the lowest-ranking criterion; and 4) repeat the process with the features not discarded until the number obtained is 1. This process can be graphically visualised and allows practitioners to evaluate the subset of features that optimizes the performance of the classifier. The final step is to determine the performance of the classifier (confusion matrix) using the subset of top-ranked features.
Results
The study included 79 recently diagnosed RRMS patients and 69 age-matched healthy control subjects. Mean age was 45.64 ± 13.59 years in the patient group and 46.94 ± 12.64 years in the control group (p = 0.604). Gender distribution was 68 women and 11 men in the patient group and 46 women and 23 men in the control group. In the MS group, mean disease duration was 1.42 ± 0.72 years and mean Expanded Disability Status Scale (EDSS, performed by EGM) score was 1.28 (range: 1–3), meaning our population presented early-stage RRMS.
Feature ranking
Figure 3 shows the accuracy against the number of features in the SVM-RFE-LOOCV method, obtained with the following hyperparameters: SVM classifier: linear kernel, C = 1.0 (regularisation parameter), penalty = L2, loss function= squared hinge. The best accuracy value (accuracy = 0.88) reported by the classifier is obtained with the seven top-ranked features shown in Table 1.
Fig. 3. Graphical representation of the accuracy to detect the disease as a function of the number of features included in the analysis.
Feature selection based on SVM-RFE-LOOCV (linear kernel).
Table 1.
Top 7 features ranked according to their weight.
| Feature | Index Ranking | Correlation with EDSS |
|---|---|---|
| Diff[Zone2-(GCL)] | 0.179 | r = 0.31, p = 7.62e-03 |
| Avg[Zone2-(IPL)] | 0.167 | r = −0.51, p = 4.33e-06 |
| Avg[Zone6-(GCL)] | 0.159 | r = −0.54, p = 1.09e-06 |
| Diff[Zone2-(IRL)] | 0.159 | r = 0.25, p = 3.43e-02 |
| Diff[Zone4-(IRL)] | 0.149 | r = 0.07, p = 5.42e-01 |
| Avg[Zone2-(GCL)] | 0.105 | r = −0.53, p = 2.27e-06 |
| Diff[Zone3-(GCL)] | 0.082 | r = 0.26, p = 3.03e-02 |
The correlation between each feature and the EDSS is indicated.
Figure 4 shows the correlation between the seven top-ranked features and the EDSS value. Among the variables that evaluate the inter-eye difference, the correlation is positive (the greater the EDSS value, the greater the thickness difference between the patients’ eyes) and among the variables that measure the average thickness, the correlation is negative (the greater the EDSS value the lesser the average thickness).
Fig. 4. Graphical representations of the linear correlation between the EDSS (Expanded Disability Status Scale) value and the seven top-ranked features.
Linear correlation between the EDSS value and the seven top-ranked features.
Automatic classifier
The classifier implemented with the seven top-ranked features obtains sensitivity = 0.86, specificity = 0.90, and accuracy = 0.88. One advantage of the SVM classifier is that it requires minimal parameter tuning, unlike deep learning classifiers that require a large training set. When evaluating the performance of a classifier, it has been shown to be important to separate training and test data to avoid optimistically biased performance estimates [14]. Sensitivity, specificity and accuracy values have been calculated from the confusion matrix and this has been obtained using leave-one-out cross-validation, where the validation data are completely disjoint from the training data. In this way, the ability of the classifier to generalise to unseen data is assessed.
Discussion
This study aims to find a biomarker capable of differentiating between patients with recently diagnosed RRMS and no history of ON and healthy subjects using only OCT, which is an innocuous, patient-friendly digital image-analysing technology with a low cost for healthcare systems.
The method implemented in this study identifies the type of variable (mean thickness value for both eyes, inter-eye difference), the retinal layer, and the areas that most influence the classification of control subjects versus recently diagnosed RRMS patients using an SVM classifier with linear kernel.
The XAI analysis showed that the key information that results in the best accuracy is found in the seven top-ranked features: Diff[Zone2-(GCL)], Avg[Zone2-(IPL)], Avg[Zone6-(GCL)], Diff[Zone2-(IRL)], Diff[Zone4-(IRL)], Avg[Zone2-(GCL)], and Diff[Zone3-(GCL)]. According to our results, Zone 2 is the most relevant (three of the five extremely relevant variables correspond to this zone). This zone corresponds to the central part of the papillomacular bundle (central PMB), which directly carries the ganglion cell axons from the macula to the brain via the optic nerve head. According to previous studies, it is the zone that is first affected in neurodegenerative pathologies [15, 16].
The results obtained in this paper, which relate to the retinal regions that undergo greatest structural alteration in RRMS, are in line with other, partly similar, research. In the diagnosis of paediatric MS patients (n = 57; age at OCT scan: 15.7 (2.2) years; disease duration at OCT scan: 0.6 (1.5) years) versus control subjects [17] combine the RFE method with a random forest classifier and conclude that the ganglion cell inner plexiform layer thickness in the supero-temporal sector and the retinal nerve fibre layer thickness in the temporal quadrant were among the most discriminant features. Hernández et al. [18] also use the PPole protocol with a Spectralis SD-OCT device; they employ RNFL and GCL layer thicknesses evaluated using the SHAP (SHapley Additive exPlanations) and Explainable Boosting Machine methods and identify Zone 1 and Zone 2 as the most relevant features in the GCL layer.
This study provides insights into the effects of early-stage RRMS on the structure of the neuroretina based on a topographical thickness map of the IRL, which is considered a continuation of the central nervous system towards the eye and is therefore affected by the disease, as shown by our correlation analysis. We observed that there is greater damage to the IRL in patients with more severe RRMS (as per the EDSS) and this finding therefore confirms the appropriateness of using OCT as a biomarker of RRMS and its progression. It also confirms its potential utility in detecting subjects with worse prognoses, as described by previous authors [19–22].
Our results highlight the importance of inter-eye difference in identifying RRMS patients with no history of ON. Moreover, the reliability of this measurement is reaffirmed by its stability over time [9]. In contrast to other papers, in our study inter-eye difference is of greater relevance than average thickness value. In light of this, it should be clarified that the numerical differences are minimal (Table 1, index ranking), that each analysis method is based on its own internal processes, and that, furthermore, the possibility that some patients have had subclinical ON cannot be ruled out.
In addition, and in line with other studies [5], our findings suggest that in order to obtain an accurate early RRMS diagnosis system it is advisable to use both inter-eye difference and the average retinal thickness measurements taken from each subject eye. While our results support other studies that show the benefits to prediction of combining measurements from several layers of the retina [23, 24], we agree that pure OCT data constitute a very useful tool that, via a simple, inexpensive, and innocuous method, can achieve even greater precision than magnetic resonance imaging. Numerous experts recommend using OCT as an additional method in standard MS diagnosis [20, 25, 26].
There are four potential limitations to this study. First, we only provide OCT measurements derived from one type of OCT device. Measurements between different devices are not identical. Thus, since the PPOLE protocol and the zones defined in Fig. 1 B were used, analysis cannot be extrapolated to other OCT devices or other protocols. Second, in MS there is the possibility that asymmetry between eyes is due to only one patient eye suffering a subclinical episode of ON, which may not have been noticed by either the patient or the doctor. The third potential limitation is related to the fact that patients were receiving various disease-modifying therapies (DMT) that can affect axonal degeneration: We did not analyse the cohort by DMTs. Fourthly, the RFE method was implemented with a single classifier, and since it is a wrapper method, different classifiers may produce different results.
Our method is applicable to a well clinically characterised patient population. More studies are needed to check if this method maintains high diagnostic efficiency if we randomly perform OCTs in patients who have not been thoroughly evaluated from a clinical perspective. In addition, other neurodegenerative pathologies such as Alzheimer’s and Parkinson’s have also been shown to be associated with a thinning of the inner layers of neuroretina similar to multiple sclerosis [15, 16], so that the tool designed can be used, with some modifications, for other neurodegenerative pathologies.
In conclusion, this study contributes to identifying both anatomical structures of the retina with pathological significance in demyelinating disorders and the type of alteration (thickness and/or inter-eye difference) produced in those disorders. Using machine learning to select the most predictive OCT features suggests the technology may support early diagnosis of MS.
Summary
What was known before
Study of retinal structure based on optical coherence tomography (OCT) data can facilitate monitorization of progression in relapsing-remitting multiple sclerosis.
Artificial intelligence can provide highly reliable diagnoses of multiple sclerosis.
Most of the studies using artificial intelligence are not explainable.
What this study adds
The Support Vector Machine plus Recursive Feature Elimination with Leave-One-Out Cross Validation (SVM-RFE-LOOCV) approach is used to find the subset of features that reduces dimensionality and optimizes the performance of the classifier.
Our study contributes to identifying both anatomical structures of the retina with pathological significance in demyelinating disorders and the type of alteration (thickness and/or inter-eye difference) produced in those disorders.
Our study demonstrates that, using machine learning to select the most predictive optical coherence tomography (OCT) features, may support early diagnosis of multiple sclerosis.
Acknowledgements
Thanks to Dr. Luis E Pablo for his help with the use of the devices.
Author contributions
FJD-M has contributed to methodology, investigation, formal analysis and writing original draft. MO has contributed to methodology, investigation, software, formal analysis, data curation and review and editing. AP has contributed to methodology and investigation. LB has contributed to conceptualisation, formal analysis, data curation, writing original draft and review and editing. EMS-M has contributed to formal analysis, data curation, and review and editing. DJ-H has contributed to methodology, investigation, software, data curation and review and editing. JMM has contributed to methodology, investigation, software, data curation and review and editing. RB has contributed to methodology, investigation, software, data curation and review and editing. EV has contributed to methodology, investigation, software, data curation and review and editing. EG-M has contributed to conceptualisation, formal analysis, data curation, writing original draft and review and editing.
Funding
This study was supported by Carlos III Health Institute grants PI17/01726, PI18/1275 and PI20/00437, by the Inflammatory Disease Network (RICORS) (RD21/0002/0050) (Carlos III Health Institute and co-funded by the European Union “NextGenerationEU/PRTR), and by project reference UAH-GP2022-2 funded by the University of Alcalá Proprietary Research Programme. The funding organisations had no role in the design or conduct of this research.
Data availability
The data collected and/or analysed during the current study are available from the corresponding author upon reasonable request.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Thompson AJ, Banwell BL, Barkhof F, Carroll WM, Coetzee T, Comi G, et al. Diagnosis of multiple sclerosis: 2017 revisions of the McDonald criteria. Lancet Neurol [Internet] 2018;17:162–73. doi: 10.1016/S1474-4422(17)30470-2. [DOI] [PubMed] [Google Scholar]
- 2.van der Vuurst de Vries RM, Mescheriakova JY, Wong YYM, Runia TF, Jafari N, Samijn JP, et al. Application of the 2017 revised Mcdonald criteria for multiple sclerosis to patients with a typical clinically isolated syndrome. JAMA Neurol. 2018;75:1392. doi: 10.1001/jamaneurol.2018.2160. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Gaitán MI, Sanchez M, Farez MF, Fiol MP, Ysrraelit MC, Solomon AJ, et al. The frequency and characteristics of multiple sclerosis misdiagnosis in Latin America: A referral center study in Buenos Aires, Argentina. Mult Scler J. 2022;28:1373–81. doi: 10.1177/13524585211067521. [DOI] [PubMed] [Google Scholar]
- 4.Petzold A, Balcer LJ, Calabresi PA, Costello F, Frohman TC, Frohman EM, et al. Retinal layer segmentation in multiple sclerosis: a systematic review and meta-analysis. Lancet Neurol [Internet] 2017;16:797–812. doi: 10.1016/S1474-4422(17)30278-8. [DOI] [PubMed] [Google Scholar]
- 5.Ortiz M, Mallen V, Boquete L, Sánchez-Morla EM, Cordón B, Vilades E, et al. Diagnosis of multiple sclerosis using optical coherence tomography supported by artificial intelligence. Mult Scler Relat Disord. 2023;74:104725. doi: 10.1016/j.msard.2023.104725. [DOI] [PubMed] [Google Scholar]
- 6.Paul F, Calabresi PA, Barkhof F, Green AJ, Kardon R, Sastre‐Garriga J, et al. Optical coherence tomography in multiple sclerosis: a 3‐year prospective multicenter study. Ann Clin Transl Neurol [Internet] 2021;8:2235–51. doi: 10.1002/acn3.51473. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Petzold A, Chua SYL, Khawaja AP, Keane PA, Khaw PT, Reisman C, et al. Retinal asymmetry in multiple sclerosis. Brain [Internet] 2021;144:224–35. doi: 10.1093/brain/awaa361. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Nij Bijvank J, Uitdehaag BMJ, Petzold A. Retinal inter-eye difference and atrophy progression in multiple sclerosis diagnostics. J Neurol Neurosurg Psychiatry [Internet] 2022;93:216–9. doi: 10.1136/jnnp-2021-327468. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Patil SA, Joseph B, Tagliani P, Sastre-Garriga J, Montalban X, Vidal-Jordana A, et al. Longitudinal stability of inter-eye differences in optical coherence tomography measures for identifying unilateral optic nerve lesions in multiple sclerosis. J Neurol Sci. 2023;449:120669. doi: 10.1016/j.jns.2023.120669. [DOI] [PubMed] [Google Scholar]
- 10.Loh HW, Ooi CP, Seoni S, Barua PD, Molinari F, Acharya UR. Application of explainable artificial intelligence for healthcare: a systematic review of the last decade (2011–2022) Comput Methods Prog Biomed. 2022;226:107161. doi: 10.1016/j.cmpb.2022.107161. [DOI] [PubMed] [Google Scholar]
- 11.Guyon I, Weston J, Barnhill S, Vapnik V. Gene selection for cancer classification using support vector machines. Mach Learn. 2002;46:389–422. doi: 10.1023/A:1012487302797. [DOI] [Google Scholar]
- 12.Chylack LT. The lens opacities classification system III. Arch Ophthalmol. 1993;111:831. doi: 10.1001/archopht.1993.01090060119035. [DOI] [PubMed] [Google Scholar]
- 13.Petzold A, Albrecht P, Balcer L, Bekkers E, Brandt AU, Calabresi PA, et al. Artificial intelligence extension of the OSCAR‐IB criteria. Ann Clin Transl Neurol. 2021;8:1528–42. doi: 10.1002/acn3.51320. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Vabalas A, Gowen E, Poliakoff E, Casson AJ. Machine learning algorithm validation with a limited sample size. PLoS One. 2019;14:e0224365. doi: 10.1371/journal.pone.0224365. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Al-Nosairy K, Horbrügger M, Schippling S, Wagner M, Haghikia A, Pawlitzki M, et al. Structure–function relationship of retinal ganglion cells in multiple sclerosis. Int J Mol Sci. 2021;22:3419. doi: 10.3390/ijms22073419. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Satue M, Obis J, Rodrigo MJ, Otin S, Fuertes MI, Vilades E, et al. Optical coherence tomography as a biomarker for diagnosis, progression, and prognosis of neurodegenerative diseases. J Ophthalmol [Internet]. 2016;2016:8503859. doi: 10.1155/2016/8503859. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Ciftci Kavaklioglu B, Erdman L, Goldenberg A, Kavaklioglu C, Alexander C, Oppermann HM, et al. Machine learning classification of multiple sclerosis in children using optical coherence tomography. Mult Scler J. 2022;28:2253–62. doi: 10.1177/13524585221112605. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Hernandez M, Ramon-Julvez U, Vilades E, Cordon B, Mayordomo E, Garcia-Martin E. Explainable artificial intelligence toward usable and trustworthy computer-aided early diagnosis of multiple sclerosis from optical coherence tomography. PLoS One. 2023;18:e0289495. doi: 10.1371/journal.pone.0289495. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Garcia-Martin E, Rodriguez-Mena D, Herrero R, Almarcegui C, Dolz I, Martin J, et al. Neuro-ophthalmologic evaluation, quality of life, and functional disability in patients with MS. Neurol [Internet] 2013;81:76–83. doi: 10.1212/WNL.0b013e318299ccd9. [DOI] [PubMed] [Google Scholar]
- 20.Martinez-Lapiscina EH, Arnow S, Wilson JA, Saidha S, Preiningerova JL, Oberwahrenbrock T, et al. Retinal thickness measured with optical coherence tomography and risk of disability worsening in multiple sclerosis: a cohort study. Lancet Neurol [Internet] 2016;15:574–84. doi: 10.1016/S1474-4422(16)00068-5. [DOI] [PubMed] [Google Scholar]
- 21.Giedraitiene N, Drukteiniene E, Kizlaitiene R, Cimbalas A, Asoklis R, Kaubrys G. Cognitive decline in multiple sclerosis is related to the progression of retinal atrophy and presence of oligoclonal bands: a 5-Year follow-up study. Front Neurol. 2021;12:678735. doi: 10.3389/fneur.2021.678735. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Frau J, Fenu G, Signori A, Coghe G, Lorefice L, Barracciu MA, et al. A cross-sectional and longitudinal study evaluating brain volumes, RNFL, and cognitive functions in MS patients and healthy controls. BMC Neurol. 2018;18:67. doi: 10.1186/s12883-018-1065-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Pérez Del Palomar A, Cegoñino J, Montolío A, Orduna E, Vilades E, Sebastián B, et al. Swept source optical coherence tomography to early detect multiple sclerosis disease. The use of machine learning techniques. PLoS One [Internet] 2019;14:e0216410. doi: 10.1371/journal.pone.0216410. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Montolío A, Cegoñino J, Garcia-Martin E, Pérez del Palomar A. Comparison of machine learning methods using spectralis OCT for diagnosis and disability progression prognosis in multiple sclerosis. Ann Biomed Eng. 2022;50:507–28. doi: 10.1007/s10439-022-02930-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Manogaran P, Hanson J, Olbert E, Egger C, Wicki C, Gerth-Kahlert C, et al. Optical coherence tomography and magnetic resonance imaging in multiple sclerosis and neuromyelitis optica spectrum disorder. Int J Mol Sci. 2016;17:1894. doi: 10.3390/ijms17111894. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Wildner P, Stasiołek M, Matysiak M. Differential diagnosis of multiple sclerosis and other inflammatory CNS diseases. Mult Scler Relat Disord. 2020;37:101452. doi: 10.1016/j.msard.2019.101452. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The data collected and/or analysed during the current study are available from the corresponding author upon reasonable request.