Reflexo pupilar à luz como biomarcador para identificação de glaucoma: avaliação comparativa de redes neurais e métodos de aprendizado de máquina

Abstract

The study of retinal ganglion cells, their photosensitivity characteristics, and their relationship with physical and cognitive processes has driven research on the pupillary reflex. Controlled by the Autonomic Nervous System (ANS), dilation (mydriasis) and contraction (miosis) are involuntary reflexes. Variations in pupil diameter may indicate physical or cognitive changes in an individual. For this reason, the pupillary reflex has been considered an important biomarker for various types of diagnoses. This study aimed to improve the automated identification of glaucoma using data from the pupillary light reflex. A comparative analysis between neural networks and classical techniques was performed to segment the pupillary signal. In addition, the performance of various data processing methods was evaluated, including filtering techniques, feature extraction, sample balancing, and feature selection, analyzing their effects on the classification process. The results show an accuracy of 73.90% in the overall classification of glaucoma, 98.10% for moderate glaucoma classification, and 98.73% for severe glaucoma, providing insights and guidelines for glaucoma screening and diagnosis through the signal derived from the pupillary light response