The Case for Local AI Development: Lessons From Computer‑Aided Detection of Tuberculosis and Silicosis in Southern...

Area of Science:

• Occupational Health
• Medical Imaging
• Artificial Intelligence

Background:

• The co-epidemic of silicosis and tuberculosis (TB) in South Africa's mining sector disproportionately affects migrant workers.
• Limited access to chest X-ray (CXR) screening exacerbates health disparities.
• Existing artificial intelligence (AI)-based computer-aided detection (CAD) systems for TB lack validation in silica-exposed populations and may contain biases.

Purpose of the Study:

• To describe challenges in developing AI CAD systems for TB and silicosis.
• To present the benefits of local public-sector development initiatives for AI CAD.
• To provide empirical evidence of AI CAD biases using local and international datasets.

Main Methods:

• Utilized a local dataset of 2000 CXRs from silica-exposed Southern African mineworkers.
• Incorporated publicly available international CXR datasets and pre-trained CAD models.
• Employed dimensionality reduction analysis and visualization techniques to assess CAD performance and biases.

Main Results:

• Local CXRs formed a distinct cluster separate from international images, indicating population-specific differences.
• Reduced image resolution disproportionately degraded silicosis detection compared to TB detection.
• Accuracy metrics alone are insufficient to measure clinical reliability and may obscure deployment failures.

Conclusions:

• Local public-sector AI CAD development is a viable alternative to externally developed systems that may exclude underserved populations.
• Addressing AI CAD deficiencies requires population-representative datasets and transparent, open-source development practices.
• Local expertise can transform AI CAD into a tool for contextually appropriate diagnostics and equitable AI deployment.