Filtragem como pré-processamento de CGAN na detecção de glaucoma

This work proposes the use of medical image pre-processing to improve the accuracy standard deviation of machine learning-based diagnostic methods for retinopathies. For this purpose, results obtained from adversarial generative neural networks conditioned for the task of retinal segmentation are used, on which the proposed filtering methods are applied in order to observe the decrease in the standard deviation of accuracy in the identification of retinal structures. retina, such as aspects of the optic nerve, veins and arteries, as well as investigations into other retinopathies, such as AMD (Age-Related Macular Degeneration) and hyposphagma. Showing an improvement in the accuracy standard deviation of 11.11% for the DSC neural network, and 18.18% for the IoU network.