Classification of Trypanosoma brucei mammalian life cycle stages using Deep Learning Algorithms

• Published in: PLoS neglected tropical diseases Vol. 19; no. 8; p. e0013298
• Main Authors: Cheraghi, Hamid, López-Escobar, Lara, Rino, José, Figueiredo, Luisa M., Szabó, Bálint
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 01.08.2025; Public Library of Science (PLoS)
• Subjects: Algorithms; Animals; Biology and Life Sciences; Computer and Information Sciences; Deep Learning; Humans; Image Processing, Computer-Assisted - methods; Life Cycle Stages; Medicine and Health Sciences; Research and Analysis Methods; Trypanosoma brucei brucei - classification; Trypanosoma brucei brucei - growth & development; Trypanosomiasis, African - parasitology
• ISSN: 1935-2735; 1935-2727; 1935-2735
• DOI: 10.1371/journal.pntd.0013298

Accurate classification of Trypanosoma brucei bloodstream forms, slender and stumpy, is essential for understanding parasite biology and transmission dynamics. Traditional classification methods rely on flourescent transgenic parasites, as distinguishing these forms visually is highly challenging. To address this, we developed a semi-automated deep-learning pipeline that segments and classifies T. brucei bloodstream forms from unlabeled microscopic images. The pipeline consists of two key stages: (1) a segmentation step using the Cellpose algorithm, which detects and extracts individual parasites while filtering out artifacts, and (2) a classification step utilizing a deep learning model based on the Xception architecture. The classification model, optimized through transfer learning and fine-tuning, achieved a 97% accuracy, outperforming standard architectures such as InceptionV3, ResNet50, and VGG16. Our results demonstrate the effectiveness of deep learning in parasite stage classification, offering a scalable and efficient approach for high-throughput analysis. Beyond T. brucei, our framework can be adapted for other single-cell classification tasks based on unlabeled morphology, contributing to advancements in biomedical imaging and automated cell analysis.