Clinical applications of artificial intelligence in identification and management of bacterial infection: Systematic...

Area of Science:

• Medical Imaging and Artificial Intelligence
• Computational Pathology
• Public Health Informatics

Background:

• Pneumonia poses a significant global public health threat, particularly to children under five and the elderly.
• Early and accurate diagnosis is crucial for managing pneumonia, especially in immunocompromised individuals, to prevent severe complications.
• Deep learning (DL) presents a promising avenue for enhancing diagnostic capabilities in medical imaging.

Purpose of the Study:

• To systematically review and meta-analyze the diagnostic accuracy of deep learning models for pneumonia detection.
• To assess the sensitivity and specificity of DL models in identifying pneumonia from radiographic images.
• To evaluate the reliability of DL models in distinguishing bacterial pneumonia from viral or fungal types.

Main Methods:

• A systematic review and meta-analysis adhering to PRISMA guidelines.
• Literature search conducted across PubMed, Scopus, and Google Scholar (January 2016 - July 2023).
• Inclusion of ten studies comprising 126,610 images and 1706 patients.

Main Results:

• Pooled sensitivity for DL models was 0.90 (95% CI: 0.85-0.94) with low heterogeneity (I2=90.20%).
• Pooled specificity was 0.89 (95% CI: 0.86-0.92) with low heterogeneity (I2=92.72%).
• Consistent and reliable estimates across studies indicate high diagnostic performance.

Conclusions:

• Deep learning models demonstrate high accuracy, sensitivity, and specificity for pneumonia diagnosis via chest X-rays.
• These models are reliable for identifying bacterial pneumonia and differentiating it from other forms.
• DL offers a valuable tool for timely pneumonia diagnosis in both pediatric and adult populations.