Convolutional neural networks-driven approach for human intestinal parasite egg recognition in microscopy images

Human intestinal parasitic diseases and intestinal-associated parasites represent an important public health concern in tropical countries. These infections are typically diagnosed through clinical evaluation combined with fecal analysis and microscopic examination, which requires highly trained lab- oratory personnel. Even so, the process remains time-consuming and prone to diagnostic errors. In this study, we propose a Convolutional Neural Network (CNN)-based approach for automated recognition of human intestinal parasite eggs in microscopy images prepared using the Kato-Katz technique. A real dataset of microscopy images from human fecal thick-smear slides was prepared, containing positive samples for five helminth egg types: hookworms ( Ancylostoma duodenale/Necator americanus ), Ascaris lumbricoides, Enterobius vermicularis, Trichuris trichiura and Schistosoma mansoni . Data augmentation techniques expanded the number of samples and, then, four CNN architectures (DenseNet, Inception, ResNet, and SqueezeNet) were evaluated for the classification task. The models, trained using transfer learning from the ImageNet, showed that DenseNet presented the best performance with 96.8% accuracy, 97.0% precision, 96.8% recall, and 96.7% F1-score. The analysis confirms that deep learning models can effectively recognize and classify parasite eggs in fecal thick-smear images, showing strong potential for integration into automated diagnostic systems to improve intestinal parasite detection. • We compare CNN models for classifying multiple human intestinal parasite eggs. • DenseNet achieved the best performance across multiple classification metrics. • Demonstrated potential integration of CNN models into clinical diagnostic workflows.