Multi-Model Ensemble Deep Learning Method to Diagnose COVID-19 Using Chest Computed Tomography Images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Deep learning enhances automated diagnosis from computed tomography (CT) scans for COVID-19.
• Ensemble learning often yields superior performance over individual models in medical image analysis.

Purpose of the Study:

• To develop and evaluate an ensemble learning method for accurate COVID-19 diagnosis using CT images.
• To compare the proposed ensemble method against single deep learning models and existing state-of-the-art approaches.

Main Methods:

• An ensemble learning framework integrating multiple neural networks was proposed.
• Two ensemble strategies were investigated: adaptive weight adjustment via backpropagation and voting.
• A dataset of 8,347 CT slices from COVID-19, pneumonia, and normal subjects was utilized for training and testing.

Main Results:

• The ensemble method achieved a high accuracy of 99.37% (recall: 0.9981, precision: 0.9893) on the training set.
• Average test accuracy reached 95.62% (recall: 0.9587, precision: 0.9559), outperforming single models by approximately 7%.
• The proposed method demonstrated up to a 10.88% accuracy improvement over the latest deep learning models on the same test set.

Conclusions:

• The developed ensemble learning method shows significant promise for the automated diagnosis of COVID-19 from CT images.
• The integration of multiple neural networks with ensemble strategies offers a robust and accurate solution for COVID-19 detection.
• This approach represents a valuable advancement in leveraging artificial intelligence for infectious disease diagnostics.