PneuNet: deep learning for COVID-19 pneumonia diagnosis on chest X-ray image analysis using Vision Transformer

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Pneumonia diagnosis, particularly subtle ground-glass lung textures, remains a challenge.
• Conventional methods and convolutional neural networks struggle with accurate feature extraction from chest X-rays (CXRs).
• The COVID-19 pandemic necessitates rapid and precise diagnostic tools for lung conditions.

Purpose of the Study:

• To develop an advanced deep learning model for accurate pneumonia diagnosis using CXR images.
• To improve the extraction and recognition of pathological lung features, especially in mild cases.
• To leverage Vision Transformer (VIT) architecture for enhanced medical image analysis.

Main Methods:

• A novel Vision Transformer (VIT)-based model, PneuNet, was developed for CXR analysis.
• The model utilizes channel-based multi-head attention on image patches, not feature patches.
• Focus is on extracting feature maps rather than just pattern recognition.

Main Results:

• PneuNet achieved 94.96% accuracy in a three-category classification task on the test set.
• The model demonstrated superior performance compared to existing deep learning models.
• Effective identification of pathological lung textures was achieved.

Conclusions:

• The proposed PneuNet model offers a highly accurate method for pneumonia diagnosis via CXR.
• VIT-based models with channel attention show significant promise for medical imaging applications.
• This approach addresses the need for improved diagnostic accuracy in lung diseases.