Parasitic egg recognition using convolution and attention network.

Area of Science:

• Medical Parasitology
• Computer Vision
• Machine Learning

Background:

• Intestinal parasitic infections (IPIs) are a major global health concern, particularly in low- and middle-income countries.
• Current methods for identifying parasitic eggs in microscopy suffer from inaccuracies and low sensitivity.
• Automated image processing offers a potential solution for improved diagnosis.

Purpose of the Study:

• To develop and evaluate a highly accurate and fast method for recognizing and classifying parasitic eggs in microscopic images.
• To address the limitations of existing diagnostic techniques for IPIs.

Main Methods:

• Utilized the Chula-ParasiteEgg dataset (11,000 images) for training and evaluation.
• Implemented and compared Convolutional Neural Network (CNN) and Convolutional Attention Network (CoAtNet) models.
• Fine-tuned the CoAtNet model specifically for microscopic images of parasitic eggs.

Main Results:

• The proposed CoAtNet model achieved an average accuracy of 93%.
• The CoAtNet model demonstrated an average F1 score of 93%.
• The fine-tuned CoAtNet model exhibited high recognition performance on the parasitic egg dataset.

Conclusions:

• The developed CoAtNet model provides a significant improvement in the accuracy and speed of parasitic egg identification.
• This solution has the potential to be integrated into automated systems for parasitological diagnosis.
• The findings pave the way for more efficient and reliable detection of IPIs.