Deep Learning Models to Predict Fatal Pneumonia Using Chest X-Ray Images

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Pneumonia Diagnostics

Background:

• Chest X-ray (CXR) is crucial for pneumonia assessment.
• Deep learning (AI) is increasingly used for medical image interpretation.

Purpose of the Study:

• To investigate the feasibility of classifying fatal pneumonia using deep learning models on CXR images.
• To evaluate the performance of AI models in differentiating fatal from nonfatal pneumonia cases.

Main Methods:

• Two deep learning models were developed using publicly available platforms.
• 1031 nonfatal and 243 fatal pneumonia CXR images were used for training and validation.
• Model performance was assessed using metrics like sensitivity, specificity, and F1 score.

Main Results:

• The first model achieved an area under the precision-recall curve of 0.929 for fatal pneumonia classification.
• External validation showed a sensitivity of 68.0% and accuracy of 77.0% for the first model.
• The second model demonstrated comparable performance, with results similar to those of respiratory physicians.

Conclusions:

• Deep learning models demonstrate good accuracy in classifying fatal pneumonia from CXRs.
• AI has the potential to assist clinicians in assessing pneumonia severity.
• Further AI performance improvements could enhance clinical decision-making for pneumonia patients.