Deep Learning for Pneumonia Detection in Chest X-ray Images: A Comprehensive Survey

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Radiology

Background:

• Pneumonia detection using chest X-rays (CXRs) is crucial for global health.
• Deep learning (DL) offers potential for automated and aided diagnosis.
• The COVID-19 pandemic highlighted the need for efficient diagnostic tools.

Purpose of the Study:

• To comprehensively analyze DL applications for pneumonia detection in CXRs.
• To evaluate DL approaches during the 2020-2023 COVID-19 period.
• To identify limitations and effectiveness of current DL methods.

Main Methods:

• Systematic literature review and analysis of DL techniques for CXR-based pneumonia detection.
• Evaluation of approaches focusing on the 2020-2023 timeframe.
• Analysis of DL models as aids or substitutes for radiologists.

Main Results:

• Vision transformers (ViTs) show significant promise for pneumonia detection in CXRs.
• Current ViT approaches face limitations including biased datasets and lack of data/code availability.
• Challenges remain in model explainability, comparison, class imbalance, and adversarial attacks.

Conclusions:

• ViTs represent the most promising DL technique for CXR pneumonia detection.
• Addressing limitations in datasets, explainability, and robustness is critical for ViT advancement.
• Further research is essential to overcome identified challenges and improve clinical utility.