South Africa flag
Case Study

South Africa – Performance of CAD4TB in a pediatric population

A radiographer reviewing a CAD4TB chest X-ray result on a monitor, with an AI-generated heat map highlighting lung abnormalities, at a health facility in South Africa.

Computer-aided detection (CAD) algorithms for tuberculosis screening are primary validated for individuals aged 15 years and older, in line with WHO guidance. Evidence on CAD performance in children remains limited. This study evaluated the diagnostic performance of CAD4Tb models for detecting radiological TB in pediatric chest X-rays.

*This study was conducted using CAD4TB, now part of CAD4TB+.

Intervention

Digital chest X-rays from children aged 0-13 years enrolled in a TB diagnostic cohort where analyzed using three CAD models, including a pediatric-adapted research version. CAD outputs were compared against a reference standard based on consensus readings from expert radiologists.

Result
  • The pediatric-adapted CAD model demonstrated higher diagnostic accuracy than standard CAD4TB versions.
  • Area-under-the-curve values increased from approximately 0.71-0.75 for standard models to 0.78 for the pediatric model.
  • At high specificity, sensitivity improved to nearly 40%, indicating better detection of TB-related radiological abnormalities in children.
Conclusion
  • The improved performance observed after limited pediatric retraining highlights the value of child-specific CAD models.
  • Further validation using larger datasets is recommended to support programmatic use in pediatric TB screening.
A radiographer reviewing a CAD4TB chest X-ray result on a monitor, with an AI-generated heat map highlighting lung abnormalities, at a health facility in South Africa.

REFERENCE: Pamer M. et al. (2024). Performance of computer-aided detection for paediatric tuberculosis screening. Conference presentation, The Union World Conference on Lung Health.