Deep learning applications in parasite microscopy: A scoping review.

Abstract

Background: Parasitic diseases remain a significant contributor to the global burden of disease and conventional microscopy remains the gold standard for parasite detection. However, it is labour-intensive and dependent on skilled personnel, leading to variability in diagnostic accuracy. Recent advances in deep learning have enabled automated analysis of microscopy images, offering potential improvements in diagnostic efficiency and consistency.

Methodology/principal findings: This scoping review aimed to map current evidence on deep learning applications in parasite microscopy for human clinical diagnosis. A systematic search of seven electronic databases was conducted for studies published between January 2015 and July 2025. Eligible studies included primary research that applied deep learning models to parasite microscopy images and reported performance metrics. A total of 118 studies were included. Image classification was the most commonly reported task, followed by object detection, while image segmentation was less frequently investigated. A wide range of architectures was used, including convolutional neural networks (CNNs), YOLO-based detection models and hybrid approaches. Most studies reported high performance across tasks, however, direct comparison was limited due to heterogeneity in datasets, evaluation metrics and experimental designs. In addition, most studies relied on internal validation using subsets of the same dataset, with limited external validation. Among the studies included in this review, several proposed implementation strategies. However, relatively few reported implementation or evaluation in real-world clinical or diagnostic laboratory settings.

Conclusions/significance: Deep learning demonstrates substantial potential for automated parasite detection in microscopy. Among the studies included in this review, most deep learning models were evaluated under experimental conditions, with relatively few reporting implementation or evaluation in clinical or diagnostic laboratory settings. Key challenges include dataset heterogeneity, lack of standardised evaluation and limited real-world validation. Future research should prioritise external validation and implementation-focused studies to support integration into clinical and laboratory workflows.