PlasmoCount 2.0: Rapid Multi-Species Malaria Parasite Detection Using Deep Learning

Visual examination of Giemsa-stained red blood cell smears is the gold-standard for identification of malaria parasite infection. Despite its wide usage, however, smear counting is time consuming, and the storage of slides has limited archiving or referencing capacity. Towards automation of smear counting and to support digital archiving, we previously developed a deep-learning application, called PlasmoCount, that provides accurate, model-assisted counting of intracellular parasites using convolutional neural networks. PlasmoCount was specifically designed for use with the human malaria parasite Plasmodium falciparum . Here, we overhaul the platform, improving its versatility and robustness. Principally, we have updated the tool so that it is capable of detecting blood-stage infections from multiple species of human-infective and rodent-infective Plasmodium parasites and have adapted it for use with both 40x and 100x objective magnifications. These augmentations broaden the distribution of input data our model can accommodate whilst maintaining its high classification accuracy (99.8%), along with modestly improving its precision with cell detection. We see significant prediction improvements on out-of-domain data, showing the adaptability of the model for real-world applications. In addition, we have substantially reduced the processing time of PlasmoCount by replacing Faster R-CNN, the original object detection model, with YOLOv8 and using batch inference for classification - modifications that reduce the latency by 90%, processing a single image in under 3 seconds (reduced from 40). Finally, we provide an offline, on-device version of the standardised framework, now referred to as PlasmoCount 2.0, which is compatible with most smartphones, including iOS and Android operating systems. PlasmoCount 2.0 markedly improves the time taken for malaria parasite smear-based detection and provides a reproducible means to assess parasite infections in laboratory settings as well as providing a roadmap for future application of the platform in clinical or field diagnosis.