COFE-Net: An ensemble strategy for Computer-Aided Detection for COVID-19.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Biomedical images provide crucial data for disease detection.
• Computer-based analysis aids medical professionals in timely diagnosis and treatment selection.
• Accurate COVID-19 screening is vital for patient management.

Purpose of the Study:

• To propose a robust deep learning ensemble framework, COFE-Net, for COVID-19 screening.
• To utilize chest X-rays (CXR) and CT scans for computer-aided detection (CADe).
• To enhance diagnostic accuracy through an efficient ensemble network.

Main Methods:

• Leveraging Transfer Learning with Convolutional Neural Networks (CNNs).
• Developing an ensemble network combining Inception V3, Inception ResNet V2, and DenseNet 201.
• Employing fuzzy logic and the Choquet fuzzy integral for combining decision scores.

Main Results:

• COFE-Net demonstrated superior performance compared to empirical ensembling.
• The fuzzy ensembling strategy dynamically refactors classifier weights based on input confidence scores.
• The method showed effectiveness across multiple datasets for COVID-19 detection.

Conclusions:

• COFE-Net offers a dynamic and effective approach to COVID-19 screening.
• The fuzzy logic integration provides an advantage over static ensembling methods.
• The proposed framework aids medical practitioners in computer-aided detection of COVID-19.