Protótipo de sistema web para apoio à decisão clínica com classificação automatizada de anormalidades em exames gastrointestinais

Abstract

The automated analysis of endoscopic images using Artificial Intelligence has increasingly been explored as a tool to support the identification of gastrointestinal abnormalities. However, much of the literature focuses on isolated classification models without integrating them into clinical management platforms. This study proposes a method that integrates a Computer-Aided Diagnosis module, based on the method by Viana (2026) for the classification of abnormalities in gastrointestinal exams, with a visualization module implemented as a web prototype for clinical management with Role-Based Access Control. The reproduced classification model achieved a mean accuracy of 95.97% ± 0.61% and a mean F1-score of 95.99% ± 0.60% using stratified 3-fold cross-validation. The integration of these modules results in a Clinical Decision Support System that provides physicians with the classification result and Grad-CAM activation maps, visually highlighting the image regions that influenced the model's prediction. These resources support clinical assessment, while the physician remains responsible for diagnostic judgment and the final interpretation of the examination. The results indicate the technical feasibility of this integration and provide a basis for future validation in real-world clinical settings.