IEViT: An enhanced vision transformer architecture for chest X-ray image classification

Area of Science:

• Artificial Intelligence
• Medical Imaging Analysis
• Deep Learning

Background:

• Chest X-rays are crucial for diagnosing conditions like pneumonia, tuberculosis, and COVID-19.
• Radiologist interpretation can be a bottleneck, particularly in underserved regions.
• Machine learning offers potential for automated chest X-ray diagnosis.

Purpose of the Study:

• To evaluate a novel Transformer-based deep learning model for chest X-ray image classification.
• To enhance the diagnostic accuracy of automated chest X-ray analysis.

Main Methods:

• Assessed the performance of the Vision Transformer (ViT) model on chest X-ray datasets.
• Developed and evaluated the Input Enhanced Vision Transformer (IEViT), a novel model.
• Tested models on four diverse chest X-ray datasets encompassing tuberculosis, pneumonia, and COVID-19.

Main Results:

• The IEViT model consistently outperformed the ViT model across all datasets.
• IEViT achieved F1-scores between 96.39% and 100%, an improvement of up to 5.82% over ViT.
• IEViT demonstrated superior sensitivity (up to +3%) and precision (up to +6.41%) compared to ViT.

Conclusions:

• The proposed IEViT model shows superior performance and generalization ability for chest X-ray classification.
• IEViT offers a powerful, cost-effective, and accessible tool to aid in medical diagnosis.
• This AI-driven approach can significantly enhance diagnostic capabilities using readily available chest X-ray imaging.