Analysis of multifocal electroretinograms from a population with type 1 diabetes using partial least squares reveals spatial and temporal distribution of changes to retinal function

Spatial–temporal partial least squares (ST-PLS) is a multivariate statistical analysis that has improved the analysis of modern imaging techniques. Multifocal electroretinograms (mfERGs) contain a large amount of data, and averaging and grouping have been used to reduce the amount of data to levels that can be handled using traditional statistical methods. In contrast, using all acquired data points, ST-PLS enables statistically rigorous testing of changes in waveform shape and in the distributed signal related to retinal function. We hypothesise that ST-PLS will improve analysis of the mfERG. Two mfERG protocols, a 103 hexagon clinical protocol and a slow-flash mfERG (sf-mfERG) protocol, were recorded from an adolescent population with type 1 diabetes and an age similar control population. The standard mfERGs were analysed using a template-fitting algorithm and the sf-mfERG using a signal-to-noise measure. The results of these traditional analysis techniques are compared with those of the ST-PLS analysis. Traditional analysis of the mfERG recordings revealed changes between groups for implicit time but not amplitude; however, the spatial location of these changes could not be identified. In contrast, ST-PLS detected significant changes between groups and displayed the spatial location of these changes on the retinal map and the temporal location within the mfERG waveforms. ST-PLS confirmed that changes to diabetic retinal function occur before the onset of clinical pathology. In addition, it revealed two distinct patterns of change depending on whether the multifocal paradigm was optimised to target outer retinal function (photoreceptors) or middle/inner retinal function (collector cells).