Inteligência artificial no combate à malária: classificação de imagens microscópicas com redes neurais convolucionais

Abstract

Malaria remains one of the major public health challenges worldwide, necessitating effective strategies for diagnosis and control. This work proposes the use of artificial intelligence, specifically Convolutional Neural Networks (CNN), for the classification of microscopic images of human cells infected by the *Plasmodium spp.* parasite. Automated classification has the potential to accelerate diagnosis and enhance accuracy in detecting infections. To achieve this objective, a public dataset containing 27,527 images was utilized, divided into healthy and infected cells. The images underwent rigorous preprocessing, including resizing, normalization, and data augmentation, thereby optimizing model training. The CNN architecture consisted of convolutional, pooling, and dropout layers, ensuring effective feature extraction. The results indicated an accuracy of 93.21%, surpassing the performance of previous models. In addition to the model, an intuitive web application was developed, allowing users to easily upload and analyze images. The challenges faced, such as the long training time (12 hours), highlight the need for more robust computational infrastructure for future implementations. This study emphasizes the role of artificial intelligence in improving malaria diagnostic methods, showcasing its applicability in public health and its potential to transform disease management through automated techniques.