A lightweight deep-learning model for parasite egg detection in microscopy images.

Area of Science:

• Medical Imaging
• Computer Vision
• Parasitology

Background:

• Intestinal parasitic infections remain a significant public health issue in developing nations.
• Accurate diagnosis relies on detecting parasite eggs in samples.
• Current automated detection methods are computationally intensive, limiting accessibility.

Purpose of the Study:

• To develop a lightweight deep learning model for efficient and accurate parasitic egg detection.
• To reduce the computational resources required for automated parasite egg identification.
• To lower the cost of automated diagnostic systems for parasitic infections.

Main Methods:

• A novel lightweight deep learning model, YAC-Net, was designed.
• YAC-Net is based on the YOLOv5n architecture with modifications to the neck (FPN to AFPN) and backbone (C3 to C2f modules).
• Experiments utilized the ICIP 2022 Challenge dataset with fivefold cross-validation.

Main Results:

• YAC-Net demonstrated improved performance over YOLOv5n, with a 1.1% increase in precision and a 2.8% increase in recall.
• The model achieved a 0.0195 higher F1 score and a 0.0271 higher mAP_0.5, while reducing parameters by 20%.
• On the test set, YAC-Net achieved 97.8% precision, 97.7% recall, 0.9773 F1 score, and 0.9913 mAP_0.5 with 1,924,302 parameters.

Conclusions:

• YAC-Net effectively optimizes model structure and simplifies parameters without compromising detection performance.
• The model reduces equipment requirements for automated detection systems.
• YAC-Net facilitates the automated detection of parasite eggs in microscope images.