Análisis avanzado de registros de electrorretinografía multifocal aplicado al diagnóstico de esclerosis múltiple

Abstract

Multiple sclerosis (MS) is a demyelinating, acquired, chronic disease that perturb the normal functioning of the white matter of the central nervous system. In a large number of cases, the visual pathway is affected during the course of MS, and even in phases prior to the disease; for this reason, the study of the visual pathway in the diagnosis is pertinent. Electrophysiological tests provide information about the functioning of the visual pathway. Among the available tests, the multifocal electroretinogram (mfERG) is a technique that allows obtaining objective and qualitative measurements of the functioning of the retina through different types of visual stimuli with a high topographic resolution. However, the number of studies concerning changes in the retina due to MS is quite limited and all of them used the classical analysis of amplitudes and latencies. The results in these studies are inconclusive. The purpose of this thesis has been to explore the capability of multifocal electroretinography for research and clinical diagnosis in patients of Multiple Sclerosis, using advanced algorithms of signal analysis. MfERG registers from 6 controls (F:M=3:3) and 10 MS patients without history of optic neuritis (F:M=7:3) were examined. This registers were obtained by Reti-Port/scan 21 device. In this thesis, the signals coming from mfERG records have been studied using various mathematical techniques that had not yet been applied in this clinical field, in order to facilitate new biomarkers for the detection of MS. These techniques are the singular spectral analysis, the sparse representation and the empirical mode decomposition. With each of them, it is possible to perform a detailed analysis of the characteristics of the mfERG signals, allowing a better characterization of the retinal response to then classify them as healthy or MS. It also proposes the use of neural networks, the use of the correlation function like classification criteria or the performance of a more detailed topographic analysis to improve its applicability. The discrimination capacity of the proposed methods has been evaluated by the area under the ROC curve: AUC. Through the analysis of the markers of amplitudes and latencies, mean values of AUC are lower than 0.6158. On the other hand after the use of the mathematical techniques described, the discrimination capacity is greatly improved: neural networks, AUC of 0.7650; singular spectral analysis, AUC of 0.8348; sparse representation, AUC of 0.7515; empirical mode decomposition, AUC of 0.8726; and topographic analysis, AUC of 0.8854. In all the proposed methods of analysis of the mfERG registers, the values of discrimination between controls and patients are higher than those obtained with the traditional technique of amplitude and latency analysis. These results suggest that the analysis of mfERG records would be applicable for the diagnosis of multiple sclerosis in its initial phases