Application of artificial neural networks for binary detection of red eye syndrome

• Autores: Torres Villanueva, Marcelino, Santos Fernández, Juan Pedro
• Formato: artículo
• Fecha de Publicación: 2026
• Institución: Universidad Privada de Tacna
• Repositorio: Revistas - Universidad Privada de Tacna
• Lenguaje: español
• OAI Identifier: oai:revistas.upt.edu.pe:article/1400
• Enlace del recurso: https://revistas.upt.edu.pe/ojs/index.php/ingenieria/article/view/1400
• Nivel de acceso: acceso abierto
• Materia: Hiperemia ocular; Aprendizaje profundo; visión computacional; clasificación binaria; ocular hyperemia; deep learning; computer vision; binary classification

Red eye syndrome is one of the most frequent reasons for consultation in primary care, and its early diagnosis is challenging due to the clinical similarity among different etiologies. In this study, a binary detection approach (“red eye” vs. “normal”) was developed and evaluated by comparing convolutional neural network (CNN) architectures, Transformer-based models, and a hybrid model. A dataset of 2,298 images reorganized into two classes was used and trained under homogeneous conditions using transfer learning and fixed hyperparameters. The experiments were conducted in Python 3.10.0 using PyTorch 2.7.1+cu118, torchvision 0.22.1+cu118, timm 1.0.17, scikit-learn 1.6.1, NumPy 1.26.4, Albumentations 2.0.8, and Matplotlib 3.8.2, on hardware equipped with an NVIDIA RTX 4060 GPU (8 GB). The results showed high performance across all evaluated models (F1 > 0.92, MCC > 0.90, and AUC ≥ 0.98). The hybrid model achieved the best overall performance (AUC = 0.996, MCC = 0.925, F1 = 0.924, and accuracy = 94.20%). McNemar’s test indicated no statistically significant differences between the hybrid model and the best-performing individual model (ResNet).