Rapid multi-species malaria parasite detection using deep learning

Abstract Visual examination of Giemsa-stained blood smears remains the gold standard for malaria parasite detection but is labour-intensive, difficult to standardise, and a major barrier to scalable digital microscopy workflows. Towards automating smear counting and supporting digital archiving, we previously developed PlasmoCount, a deep learning application for accurate, model-assisted counting of intracellular parasites. Here we present PlasmoCount 2.0, a comprehensively redesigned platform that enables robust analysis across parasite species, imaging magnifications and sample preparations. Replacing Faster R-CNN with YOLOv8 and introducing batch inference we have reduced processing bottlenecks by 90%, enabling analysis of single images in under three seconds while maintaining high classification accuracy. Furthermore, by training on diverse multi-species datasets, we have improved classification performance across included species and enhanced generalisation to previously unseen Plasmodium species. Finally, the platform distinguishes white blood cells from infected erythrocytes, increasing robustness to whole-blood smears, and is deployed as an offline smartphone-compatible application that performs on-device inference without network connectivity. Together, these advances have the potential to transform automated malaria smear analysis from a specialised deep learning application into a practical, standardised platform for routine laboratory research and establish a foundation adaptable for future clinical and field deployment, including remote areas with low connectivity.