Capturing correlations in vision parameters by artificial neural networks

Understanding the correlations between different visual functions is fundamental to ensure the right intervention to solve refractive problems. Existing studies are based on complex procedures, where it is possible to address analysis through empirical models. A different approach that has shown great functionality are neural networks, since they are able to find correlations within large collections of data with little assistance from the user. The present study stems from the desire to develop a prediction algorithm through Neural Networks by adopting a black-box approach. The objective is to design a network able to predict the values of visual acuity from the refractive errors and take into account correlation between the two quantities. For the construction of the network in questions, 136 eyes (68 subjects) were included, and, even using commercial, unsophisticated tools, the predictions for visual acuity remain close to the actual one. Our results give hope for the wide diffusion of this contact between optometry and neural networks.