C2BNet: A Deep Learning Architecture With Coupled Composite Backbone for Parasitic Egg Detection in Microscopic...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Parasitology

Background:

• The Internet of Medical Things (IoMT) and Artificial Intelligence (AI) enable automated diagnosis of chronic diseases like parasitic infections using 2D microscopic images.
• Microscopic image analysis for parasitic egg detection faces challenges including focus failure, motion blur, and variable zoom levels, impacting model performance.

Purpose of the Study:

• To propose a novel deep learning model, the Coupled Composite Backbone Network (C2BNet), for accurate detection of parasitic eggs in 2D microscopic images.
• To enhance object detection performance by addressing the specific challenges inherent in microscopic imaging.

Main Methods:

• Developed C2BNet, featuring a two-path heterogeneous backbone structure to learn object features from multiple perspectives.
• Introduced a novel feature composition method to enhance feature representation across different backbone paths.
• Implemented multiscale weighted box fusion (WBF) to refine bounding box predictions for improved detection accuracy.

Main Results:

• C2BNet demonstrated satisfactory performance compared to state-of-the-art methods on the Chula-ParasiteEgg-11 dataset.
• The model effectively learned detailed morphology and semantic features, leading to more precise parasitic egg detection.
• C2BNet showed improved focus on learning intricate feature details crucial for accurate identification.

Conclusions:

• C2BNet offers a robust solution for automated parasitic egg detection in microscopic images, addressing key imaging challenges.
• The proposed architecture and fusion strategy enhance the precision of AI-driven diagnostic tools in parasitology.
• This advancement contributes to improved chronic disease management through IoMT and AI integration.