Published as the editor’s choice in the Journal of Clinical Microbiology, October 2025
A new milestone in AI-driven parasitology
Building on prior success in digitizing and automating trichrome stain analysis, this new study marks a major advancement in stool parasite detection. Researchers from ARUP Laboratories and Techcyte developed and clinically validated a deep convolutional neural network (CNN) for detecting protozoan and helminth parasites in concentrated wet mounts: a long-standing challenge in digital parasitology.
Study overview
The team trained the model on 4,049 parasite-positive specimens collected from laboratories across the United States, Europe, Africa, and Asia, encompassing 27 species of gastrointestinal parasites. The dataset represented a wide range of fixatives and preparation techniques, making it one of the most comprehensive training collections ever assembled in parasitology.
Slides were scanned using multiple whole-slide imaging systems (Pramana, Hamamatsu, and Grundium) and mounted using a custom iodine-glycerol solution that enhanced protozoan contrast while preserving helminth visibility. Training and labeling were performed using Techcyte’s cloud-based AI platform.
Model development and validation
The CNN, based on the YOLO architecture, was optimized for fine detection of both small protozoans and large helminth eggs. Validation was performed on independent datasets scanned with the Pramana SpectralHT2 system.
Key results included:
- Initial positive agreement: 94.3% (250/265 positive specimens)
- Initial negative agreement: 94.0% (94/100 negatives)
- After discrepancy resolution and expert review, positive agreement rose to 98.9% across 25 organism classes, with confirmation of 193 additional true-positive detections that were not found by manual microscopy.
The AI system also showed greater analytical sensitivity than experienced technologists, detecting organisms at lower concentrations in limit-of-detection studies for Entamoeba, Ascaris, Trichuris, and hookworm.
Global collaboration and diversity
To ensure broad clinical relevance, specimens and expertise were sourced from institutions and researchers across the United States, Europe, Africa, Asia, and Australia, including:
- ARUP Laboratories (USA)
- Mayo Medical Laboratories (USA)
- The Centers for Disease Control and Prevention (USA)
- University of the Philippines Manila (Philippines)
- Libreville University (Gabon)
- Southern Cross University (Australia)
- Catholic University of Health and Allied Sciences (Tanzania)
- Universidad Complutense de Madrid (Spain)
- Usmanu Danfodiyo University (Nigeria)
This global collaboration enabled inclusion of diverse fixatives, backgrounds, and preparation techniques, strengthening the model’s generalizability and ensuring performance across a wide range of laboratory conditions.
Implications for laboratory practice
Wet mount microscopy has remained largely unchanged for a century, relying on intensive manual review and expert interpretation. This study demonstrates that AI-assisted wet mount screening can:
- Increase diagnostic yield, revealing mixed infections and low-prevalence organisms
- Give parasitologists more time to dedicate to complex identifications and confirmations by automating the most time-consuming portions of screening
- Simplify workflows through digital image review rather than microscope screening
- Improve consistency and reproducibility across technologists
Importantly, the authors note that AI serves as a presumptive detection tool, with humans retaining ultimate verification, allowing laboratories to apply expertise more efficiently and consistently.
Conclusion
This landmark study represents the first comprehensive AI model for wet mount parasite detection, completing the transition of the ova-and-parasite exam into a fully digital, AI-assisted workflow. By pairing advanced imaging with deep learning, laboratories can modernize parasitology workflows while maintaining diagnostic accuracy and expanding global accessibility.
References
Mathison BA, Knight K, Potts J, et al. Detection of protozoan and helminth parasites in concentrated wet mounts of stool using a deep convolutional neural network. J Clin Microbiol. 2025; DOI: 10.1128/jcm.01062-25