Imagerie fonctionnelle de la rétine par électrorétinographie multi-angulaire

Abstract

Purpose: The purpose of this project is to develop a recording method of the electrophysiology of the eye allowing functional cartography of the retina. The long term objective of this approach is to provide the clinician with an imaging tool helping to diagnose certain retina pathologies (retinopathies). Context: Full field ERG is a well-known method to objectively assess retinal function. However, this method cannot express local variations in retinal response unless a specific stimulus is used, as is done in mfERG. Our solution was twofold. First, we proposed a method for recording an ERG with multiple spatial derivations (as is done in ECG). Second, we solved the electromagnetic problem of computing retinal activity from few corneal or skin potentials. Methods: We proposed a novel method that uses the rotating capability of the human eye in order to record corneal or skin potentials from various positions (i.e. the maERG). This method implies that 3 skin electrodes are positioned on the subject’s lower lid, internal canthus and external canthus. We recorded the maERG from 2 healthy subjects using 11 different gaze positions for a total of 33 virtual electrodes per subject. Based on realistic eye and skin conductivities and geometries, we built a Boundary Elements Model (BEM) of the human eye. In order to determine if the maERG method can be used for retinal imaging, we compared the performance of multiple electrodes configurations of maERG model with a model containing an electrode array in direct contact with the cornea. We simulated 2 scenarios: a central inactivity and a peripheral inactivity. We propagated the simulated retinal activity on the electrodes (i.e. the forward model). Based on these simulated electrode measurements, we reconstructed the retinal activity (i.e. the inverse problem) and compared the reconstructed image with the theoretical sources configuration using a balanced Area Under ROC-curve (AUC) and Matthews Correlation Coefficient (MCC) approaches. We compared AUCs ans MCCs obtained with all models as a metric of the maERG reconstruction potential. We also proposed a method to overcome the necessity of a high Signal-to-Noise Ratio (SNR) signals by locating shift invariant wavelet transform coefficients. Finally, to qualitatively compare our simulated result with experimental data, we reconstructed the retinal activity of healthy subjects. Results: The different skin electrodes gave different ERG signals according to each gaze position, meaning that our recording method can give multiple ERG derivations and thus, the inverse problem is feasible. Our simulations results showed that fair reconstruction (AUC>0.7 and MCC>0.5) was possible with only horizontal gaze variations, 55dB Signal-to-Noise Ratio (SNR) and a pathology surface of at least 5%. We also demonstrated that localising with wavelet coefficients, we can reduce the minimum SNR to about 30-40dB. The reconstruction of retinal activity from experimental data of healthy subjects were qualitatively similar to the healthy subject simulation meaning that our simulated and experimental data are concordant. Conclusion: We proposed a novel method for the recording of multiple derivations of the ERG response by using the rotating capability of the human eye, a method that we named the maERG. We also propose a model of the human eye and a method to solve the inverse problem in order to represent the retinal activity based on each parcel’s position, therefore generating a functional imaging of retinal responses. Even if our technique necessitates a high SNR, we demonstrated that denoising algorithms exist which would permit to achieve that signal quality and thus, good retinal mapping resolution. We believe that our novel technique of recording the ERG will increase the sensitivity of the ERG thus permitting earlier diagnosis and more precise monitoring of retinopathies.

Date	30 Mar 2016
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Jean-Marc Lina (Supervisor) & Pierre Lachapelle (Co-supervisor)