Propostas de algoritmos de inteligência artificial para screening de edema macular diabético

Abstract

Introduction: Diabetic Macular Edema (EMD), a swelling in the central area of the retina resulting from vascular occlusion, is one of the consequences of diabetic retinopathy (RD). However, current tools used for its diagnosis, such as Optical Coherence Tomography (OCT), which provides data to aid in the diagnosis of macular diseases and glaucoma, require significant medical expertise to confirm phenotypes. Therefore, Artificial Intelligence (IA) tools to support decisionmaking in such a context are naturally welcome. Objectives: Thus, the intention of the author of this work is to describe a new strategy based on Artificial Intelligence (IA) to support the detection and pre-diagnosis of EMD, analyzing features originating from Optical Coherence Tomography (OCT) using Paraconsistent Feature Engineering (PFE). Materials and Methods: Subsets of 21 features described in the OCT setting Macula Map A X-Y were analyzed, in addition to visual acuity, diagnostic parameter, age, indication of the eye under analysis and gender, totaling 26 features, in a total of 700 exams, divided into two scenarios, with two or six classes, that is, different phenotypes. Specifically, based on the EFP and the comparison of the Support Vector Machine (SVM), K-Nearest Neighbourhoods (KNN), Decision Trees (DTREE) and Logistic Regression (LR), to formulate a strategy that aims to help clinicians in the pre-diagnosis of EMD. Results: The PFE-based analysis effectively found dominant representative features that allowed relevant classification accuracies based on SVM. Conclusion: The results obtained prove the possibility of using PFE in the intelligent automatic detection of EMD, due to the selection provided in the features by PFE, providing a contribution to the state-of-the-art in the area.