MULTIMODAL ARTIFICIAL INTELLIGENCE IN PREDICTING FUNCTIONAL PROGRESSION IN OPEN-ANGLE GLAUCOMA

Abstract

Introduction: Open-angle glaucoma is recognized as a leading cause of irreversible blindness worldwide, characterized by the progressive loss of retinal ganglion cells and structural changes to the optic nerve associated with functional visual field deficits. Predicting functional progression has posed a clinical challenge due to the variability of the disease's course and the need to integrate multiple risk factors—such as intraocular pressure, retinal nerve fiber layer thickness, optic disc characteristics, and patterns of visual field alteration. In this context, multimodal artificial intelligence has emerged as an innovative approach by combining diverse ophthalmic data sources—including optical coherence tomography images, fundus photographs, functional tests, and clinical information—thereby enabling the identification of complex patterns associated with glaucoma progression. Scientific studies have demonstrated that deep learning models possess an enhanced capacity to predict functional deterioration, contributing to earlier diagnosis, risk stratification, and personalized therapeutic monitoring. This systematic literature review aimed to analyze scientific evidence regarding the application of multimodal artificial intelligence in predicting functional progression in patients with open-angle glaucoma, highlighting advancements, clinical applications, and contributions to monitoring disease progression. Methodology: The review was conducted in accordance with the PRISMA checklist, utilizing scientific articles published over the last 10 years in the PubMed, SciELO, and Web of Science databases. The following keywords were used: "artificial intelligence," "multimodal deep learning," "open-angle glaucoma," "functional progression," and "visual field prediction." Original studies published in scientific journals that evaluated artificial intelligence models applied to predicting functional glaucoma progression were included. Duplicate articles, studies that did not evaluate patients with open-angle glaucoma, and works that did not analyze outcomes related to functional progression were excluded. Results: The studies analyzed demonstrated that multimodal algorithms exhibited higher predictive accuracy compared to isolated assessments, primarily due to the integration of structural and functional data. Research highlighted the use of deep neural networks to identify early patterns of visual loss, estimate the rate of progression, and assist in clinical decision-making. Furthermore, the models showed potential to reduce diagnostic delays and improve the selection of patients at higher risk of progressing to significant visual impairment. Conclusion: Scientific evidence indicated that multimodal artificial intelligence represents a promising tool for predicting the functional progression of open-angle glaucoma, enabling more individualized and efficient approaches. Integrating different examination modalities enhanced the capacity to interpret ophthalmic data, facilitating early interventions and targeted strategies for vision preservation.