Automated real-time surveillance of Bithynia snails using a comparative YOLO based approach for liver fluke host detection
1
Issued Date
2026-12-01
Resource Type
eISSN
20452322
Scopus ID
2-s2.0-105038955151
Pubmed ID
41876565
Journal Title
Scientific Reports
Volume
16
Issue
1
Rights Holder(s)
SCOPUS
Bibliographic Citation
Scientific Reports Vol.16 No.1 (2026)
Suggested Citation
Jenwithee T., Meererksom T., Limpanont Y., Sripa B., Laha T., Suwannatrai A.T. Automated real-time surveillance of Bithynia snails using a comparative YOLO based approach for liver fluke host detection. Scientific Reports Vol.16 No.1 (2026). doi:10.1038/s41598-026-43387-x Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/116863
Title
Automated real-time surveillance of Bithynia snails using a comparative YOLO based approach for liver fluke host detection
Corresponding Author(s)
Other Contributor(s)
Abstract
Bithynia species serve as obligate intermediate hosts for Opisthorchis viverrini, a Group 1 carcinogen endemic to Southeast Asia and the primary etiological agent of cholangiocarcinoma. Accurate identification of morphologically similar Bithynia species is critical for effective disease surveillance. This study systematically evaluates four You Only Look Once (YOLO) models—YOLOv5, YOLOv8, YOLOv10, and YOLOv11—for automated detection and classification of Bithynia species. Models were trained on 4,204 images containing 8,559 annotated specimens of Bithynia funiculata, B. siamensis goniomphalos, and B. s. siamensis. Model performance was assessed through comparative analyses of YOLO architectures and direct benchmarking against five human experts. YOLOv10 demonstrated the highest classification accuracy (98.7%), robust performance across diverse environmental conditions, and computational efficiency (model size: 31.9 MB; processing speed: 4.54 FPS). Our analysis reveals a fundamental complementarity between AI and human expertise: humans achieved perfect detection accuracy, while the artificial intelligence model provided superior classification of detected specimens (74.5% for YOLOv10 and 48.3% for human experts, respectively). These findings indicate that integrating human detection with automated classification in hybrid human–AI workflows could substantially improve diagnostic accuracy and throughput. This study establishes a technological foundation for scalable surveillance systems to support control of O. viverrini transmission in resource-limited endemic regions.