Diseño de un sistema basado en redes neuronales convolucionales para la detección y clasificación automatizada de quemaduras

Abstract

Burns represent a public health problem with a high incidence and risk of serious complications if not diagnosed and treated promptly. Conventional clinical diagnosis, based on visual inspection, presents a considerable margin of error, especially in settings without specialized personnel. In this context, an automated diagnostic system for the detection and classification of burns by degree, using convolutional neural networks, was developed. The proposed approach employs pre-trained ResNet50 architecture, optimized through transfer learning and fine-tuning, using a curated dataset of clinical burn images obtained from public repositories. The preprocessing pipeline included image normalization, region-of-interest cropping, and data augmentation techniques to enhance the model’s robustness. Performance evaluation was conducted using accuracy, precision, recall, specificity, and area under the ROC curve (AUC), complemented by confusion matrix analysis and class-wise ROC curves. Experimental results demonstrate that the ResNet50-based model achieved consistent multiclass classification performance, with strong discriminative ability across burn severity levels. Furthermore, a web platform integrating the trained model was developed, enabling users to upload images and receive rapid, accessible preliminary diagnoses. This research highlights the potential of artificial intelligence as a complementary tool in medical diagnostics, particularly in scenarios with limited specialist availability.