Signal identification for visual electrophysiological recordings. A comparison of signal analysis techniques and their application to clinical electrophysiology of vision.

Abstract

Aims: The aim of this thesis was to investigate the use of objective methods for the analysis of visual electrophysiological recordings. Specifically can signal identification algorithms identify electrophysiological signals and can they be applied to improve clinical testing and analysis? Methods: Automated signal identification algorithms were applied to multifocal electroretinogram (mfERG) and visual evoked potential (VEP) recordings. To simulate the types of signal identification problems encountered in the clinical environment, recordings were performed on healthy volunteers then artificially modified to represent the effects of disease. A multivariate analysis, spatial-temporal partial least squares (st-PLS) was applied to mfERGs recorded from a population of patients with Type 1 diabetes. Results: Signal identification algorithms were able to identify mfERG and VEP responses that had been artificially attenuated. The best performing algorithms outperformed human expert observers at identifying preserved mfERG responses. Application of signal detection algorithms increased the quality and reduced the time for recording VEPs. Metrics of algorithm performance demonstrated that algorithms using more prior knowledge about expected waveform morphology performed better than algorithms that were naive. Changes to retinal function in patients with Type 1 diabetes, measured using the mfERG, were detected using st-PLS analysis. The st-PLS analysis revealed information about the spatial and temporal distribution of these changes that was not revealed using traditional analysis methods. Conclusions: The application of more advanced analytical techniques can increase the accuracy and decrease the time required for clinical testing. Multivariate analysis techniques can reveal novel information about disease etiology.