Automated pathogen identification system for UNEFM microbiology laboratories

Abstract

This undergraduate thesis proposes the development of an automated pathogen identification system for the microbiology laboratories of UNEFM, integrating computer vision with the ESP32-CAM module, convolutional neural networks trained with local samples, and a web platform hosted on an ESP32. Its main objective is to optimize microbiological diagnosis, reducing response time from over 72 hours to less than 5 minutes, minimizing human errors, and ensuring clinical traceability through a local database and PDF report generation. The methodology included a situational diagnosis, technical, operational, and economic feasibility studies, and prototyping with low-cost hardware and AI-based software. The obtained results demonstrate an overall accuracy of 93.8% in classifying six classes of pathogens, an average inference time of 3.3 seconds, and a functional web interface that allows patient registration, history consultation, and clinical report generation. Hardware and communication tests, CNN model accuracy tests, web interface functionalities, and data management tests were executed, confirming the technical and operational feasibility of the system. It is concluded that the prototype constitutes an accessible, accurate, and traceable alternative for pathogen identification in Venezuelan public laboratories.