Image recognition based on deep learning in Haemonchus contortus motility assays

Area of Science:

• Veterinary Parasitology
• Bioinformatics
• Drug Discovery

Background:

• Anthelmintic drug resistance in parasitic nematodes like *Haemonchus contortus* necessitates novel drug discovery.
• Current drug screening relies heavily on motility assays, with a need for improved analytical methods.

Purpose of the Study:

• To evaluate the performance of a deep learning approach, Mask R-CNN, for analyzing nematode motility videos.
• To compare Mask R-CNN against existing algorithms for worm detection and motility forecasting.

Main Methods:

• Application of Mask R-CNN, a deep learning model, for analyzing *Haemonchus contortus* motility videos.
• Comparison with Wiggle Index and Wide Field-of-View Nematode Tracking Platform algorithms.
• Utilizing intersect over union for classifying motile/non-motile instances.

Main Results:

• Mask R-CNN demonstrated superior performance in worm detection and motility forecasting compared to other methods.
• Achieved a mean absolute percentage error of 7.6% for detection and 5.6% for motility forecasts.
• Intersect over union achieved 89% accuracy in classifying motile/non-motile instances.

Conclusions:

• Mask R-CNN offers a more precise and viable alternative for video analysis of worm motility in drug screening.
• The method addresses limitations in detecting overlapping objects, improving overall assay precision.
• This deep learning approach has the potential to expand video analysis techniques in parasitology research.