Capturing correlations in vision parameters by artificial neural networks.

In the analysis of vision, functions are often assessed as independent abilities, despite the fact that correlations are actually present. A more comprehensive view may assist practitioners in evaluating critically the outcomes of their measurements, as well as in highlighting possible anomalies. Due to the natural variability of physiological data, these correlations are not expected to be sharp, and a model guiding our intuition can often be hard to find. Machine learning (ML) techniques offer these days the possibility of inspecting large sets of data, and of extracting relevant information in a model-independent way. We report on the application of a prominent ML paradigm, artificial neural networks to successfully recover the correlation present between visual acuity (VA) and refractive errors with a simple algorithm. The optimisation of our network demonstrates that this operation can be performed with reduced complexity, despite the small size of our data sample. With the inception of these techniques beyond this well-known case, a path can be opened up in which more general correlations can be unveiled.