Deep learning-based detection and viability assessment of Eimeria oocysts

Area of Science:

• Veterinary Parasitology
• Computational Biology
• Poultry Science

Background:

• Coccidiosis, caused by Eimeria species, significantly impacts the global poultry industry, leading to economic losses.
• Accurate quantification of viable Eimeria oocysts is essential for effective poultry vaccination strategies.
• Current oocyst viability assessment methods are complex and unsuitable for routine monitoring.

Purpose of the Study:

• To develop a simple, cost-effective deep learning approach for distinguishing viable from non-viable Eimeria oocysts.
• To utilize morphological features, specifically granular structures in dead oocysts, for viability detection.
• To enhance coccidiosis vaccine formulation and management through reliable oocyst viability assessment.

Main Methods:

• Employed deep convolutional neural networks (YOLOv7 architecture) for viability detection.
• Utilized phase-contrast (PC), differential interference contrast (DIC), and brightfield (BF) imaging of Eimeria acervulina oocysts.
• Refined datasets with class-specific labeling (sporulated, unsporulated, dead) and performed cross-species evaluations.

Main Results:

• The YOLOv7 model trained with PC images achieved high precision (93.1%) and recall (91.2%).
• Dataset refinement significantly improved performance, reaching 99.1% precision and 99.1% recall.
• The model demonstrated excellent generalization to Eimeria tenella (100% precision, 98.1% recall) and good performance for Eimeria maxima after fine-tuning.

Conclusions:

• Deep learning offers a practical, rapid, and reliable method for Eimeria oocyst viability assessment.
• This approach can significantly improve vaccine formulation and coccidiosis management in the poultry industry.
• The methodology shows potential for assessing viability in related parasites affecting human health and food safety.