Computer-aided reading of chest radiographs for paediatric tuberculosis: current status and future directions

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Paediatric Infectious Diseases

Background:

• Computer-aided detection (CAD) systems are approved for adult tuberculosis triage via chest x-rays (CXRs), but not for children.
• Chest x-rays (CXRs) are crucial for paediatric tuberculosis evaluation in primary care, yet lack specialized automated detection tools.
• Existing CAD systems do not account for unique paediatric radiographic and clinical features.

Purpose of the Study:

• To highlight the need for paediatric-specific CAD models for tuberculosis screening and diagnosis.
• To emphasize the requirement for large, diverse paediatric CXR datasets for model development.
• To advance the application of CAD in improving equity and reducing the global burden of childhood tuberculosis.

Main Methods:

• Review of existing CAD approaches for paediatric CXR analysis.
• Discussion of the necessity for curated datasets linking paediatric CXRs with standardized tuberculosis classifications.
• Exploration of transfer learning techniques using deep learning models for paediatric applications.

Main Results:

• Current CAD systems are not optimized for paediatric chest x-rays (CXRs).
• Significant gaps exist in large, standardized datasets for paediatric tuberculosis.
• Transfer learning shows promise for developing paediatric-specific CAD models.

Conclusions:

• Paediatric-specific CAD models are essential for accurate tuberculosis screening and diagnosis in children.
• Development requires diverse, well-annotated paediatric CXR datasets.
• Advancing CAD for paediatric tuberculosis is critical for global health equity.