Computational Intelligence-Based Method for Automated Identification of COVID-19 and Pneumonia by Utilizing CXR...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Chest X-ray (CXR) interpretation for lung diseases like COVID-19 is challenging and time-consuming for radiologists.
• Automated AI methods for CXR analysis show promise for early disease detection.
• Leveraging deep learning and pre-trained models can enhance diagnostic accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate an AI-based method for automated classification of CXR images.
• To improve the accuracy and speed of identifying viral infections, specifically COVID-19, pneumonia, and differentiating between bacterial and viral pneumonia.
• To explore the effectiveness of combining pre-trained deep learning models for enhanced diagnostic performance.

Main Methods:

• Utilized transfer learning by integrating pre-trained VGG16 and InceptionV3 models.
• Developed a base model using VGG16 and InceptionV3 weights.
• Added a fully connected layer for binary and multi-class classification.
• Implemented a weight fusion technique combining layers from both models.
• Trained and tested the model on a dataset including healthy, COVID-19, viral pneumonia, and bacterial pneumonia CXR images.

Main Results:

• The proposed weight fusion method achieved 99.5% accuracy in binary classification (COVID-19 vs. healthy) within 20 epochs.
• The model attained 98.2% accuracy in three-class classification (healthy, COVID-19, pneumonia) over 100 epochs.
• The weight fusion approach demonstrated superior performance compared to existing models.

Conclusions:

• The developed AI model, utilizing a weight fusion technique with VGG16 and InceptionV3, is highly effective for automated CXR analysis.
• This method significantly improves diagnostic accuracy for COVID-19 and other lung conditions.
• The approach offers a promising solution for faster and more reliable early detection of lung diseases using medical imaging.