Enhanced Pneumonia Detection in Chest X-Rays Using Hybrid Convolutional and Vision Transformer Networks

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Deep Learning for Diagnostics

Background:

• Pneumonia detection from chest X-rays is crucial for timely treatment.
• Existing diagnostic methods face limitations, especially in resource-constrained areas.
• Deep learning offers promising avenues for automated and accurate image analysis.

Purpose of the Study:

• To develop and evaluate a novel hybrid deep learning model for enhanced pneumonia detection in chest X-rays.
• To improve diagnostic accuracy and reduce misclassifications compared to traditional methods.
• To create a robust and deployable solution for regions with limited access to healthcare.

Main Methods:

• A hybrid model integrating Convolutional Neural Networks (CNNs) with modified Swin Transformer blocks was developed.
• CNN layers extracted local features, while Swin Transformers captured global context via window-based self-attention.
• Image preprocessing included resizing and CLAHE; data augmentation and Bayesian optimization (Optuna) were used for robustness and fine-tuning.

Main Results:

• The hybrid model achieved 98.72% accuracy and a low loss of 0.064 on an unseen dataset.
• It significantly outperformed a baseline CNN model across all metrics, including precision (0.9738 normal, 1.0000 pneumonia) and F1-score (0.9872).
• Confusion matrices indicated high sensitivity and specificity, with minimal misclassifications.

Conclusions:

• The hybrid CNN-ViT model effectively captures both local and global features in X-rays, leading to superior pneumonia detection performance.
• Its lightweight design facilitates deployment in resource-limited settings, potentially improving patient outcomes globally.
• Future work includes model refinement, advanced image processing, and explainable AI integration.