Inteligência artificial aplicada ao monitoramento de fluido na degeneração macular relacionada à idade forma neovascular

Abstract

Introduction: Age-related macular degeneration (AMD) is one of the leading causes of blindness worldwide. Optical Coherence Tomography (OCT) has emerged as a central tool in AMD management, allowing visualization of key biomarkers such as retinal fluids, ellipsoid zone integrity, and retinal pigment epithelium. Recently, artificial intelligence (AI) has brought significant advancements in diagnostic accuracy, disease progression prediction, and personalized treatment approaches. Objectives: This study aimed to evaluate the use of AI algorithms to quantify retinal biomarkers in patients with neovascular AMD and analyze the correlation between fluid volumes in each compartment and visual and anatomical outcomes in a European cohort and a Brazilian cohort. Methods: In the first study, automated OCT image analysis was performed using an AI algorithm (Fluid monitor, RetInsight, Vienna) to quantify intraretinal (IRF), subretinal (SRF) fluids, and retinal pigment epithelium detachment (PED), and a machine learning-based model was developed to predict the need for treatment, visual acuity (VA), and anatomical outcomes. In the second study, automated AI-based quantification of retinal fluids was conducted, and automated segmentation of the ellipsoid zone (EZ) was performed using an ensemble U-net convolutional network, with the purpose of correlating the impact of baseline volumes of retinal fluids and PED with the loss of photoreceptor integrity. In the third study, the same algorithm for retinal fluid quantification was used in a Brazilian cohort, and baseline volumes were correlated with visual and anatomical outcomes after 2 years. Statistical analyses used predictive models, linear regression models, and longitudinal regression to evaluate the correlation between fluids in each compartment with VA, EZ integrity, and progression to macular atrophy (MA) and subretinal fibrosis (SF). Results: The results of the three studies showed that AI-based OCT biomarker monitoring tools were effective in monitoring AMD patients in both the European and Brazilian cohorts. With the used algorithms, it was possible to observe a correlation between elevated IRF and PED volumes with thinning and loss of EZ and poorer visual outcomes, as well as greater progression to MA and SF. A correlation was also observed between SRF and better visual outcomes, and lower MA progression. However, SRF showed a correlation with an increased need for intravitreal anti-angiogenic injections. Conclusions: AI-based algorithms proved promising in the monitoring of neovascular AMD, with potential to predict outcomes, refine diagnosis, and support personalized approaches for these patients.