Inside South Africa’s Mponeng Gold Mine, nearly four kilometres below the surface, a clinical artificial intelligence system analysed a miner’s chest X‑ray for tuberculosis and occupational lung disease in under 45 seconds — and crucially, did so without relying on cloud connectivity or an on‑site radiologist. The demonstration, reported by FORBES AFRICA, was led by Pretoria‑based health technology company Nexus Intelligence.
Why the test matters
The mining environment is one of the most challenging settings imaginable for medical technology: connectivity is unreliable, logistics are complex and workers face elevated risk of lung disease. Nexus co‑founder Dr Gerhard Ferreira described the site as a way to stress‑test the platform’s ability to operate “at the edge, close to the worker, in a setting with constrained connectivity, complex logistics and a high need for occupational lung surveillance.”
“Mponeng demonstrated that offline capability is real in the single most hostile environment we could find,” said Dr Gerhard Ferreira.
The system, called Nexus AI CXR, is a regulated clinical decision‑support tool that screens chest X‑rays for tuberculosis and other significant lung abnormalities including silicosis. Nexus licensed a foundational chest X‑ray model developed by Google Research and built its product to integrate with existing X‑ray workflows in facilities where radiologists are scarce or where backlogs limit timely diagnosis.
Implications for occupational health and rural care
The demonstration at Mponeng highlights two interlocking problems in South African and broader African healthcare: a shortage of specialists who can interpret radiology images, and the logistical barriers that prevent rapid diagnosis where it is most needed. Nexus’ approach is not to replace specialists but to provide decision support where radiologists are unavailable or overwhelmed.
- Edge operation: the AI runs without cloud access, addressing connectivity limitations common in mines and remote clinics.
- Speed: the system returned results in under 45 seconds during the mine test.
- Target conditions: it screens for tuberculosis and occupational lung diseases such as silicosis.
Taken together, these characteristics suggest the technology could be especially useful for occupational lung surveillance programmes in the mining sector and for primary care facilities in rural and peri‑urban areas where radiology capacity is limited.
What the technology does and does not do
Nexus markets its product as a clinical decision‑support system, not a stand‑alone diagnostic. That distinction matters for how the tool would be used in practice: it is designed to flag abnormal images and assist clinicians, rather than to be the final arbiter of disease. The company’s use of a licensed model from Google Research provides a technical foundation while Nexus builds workflow and regulatory controls around it.
| Feature | Reported detail |
|---|---|
| Environment tested | Mponeng Gold Mine, nearly 4 km underground |
| Result time | Under 45 seconds |
| Conditions screened | Tuberculosis and occupational lung abnormalities including silicosis |
| Model base | Licensed foundational chest X‑ray model from Google Research |
The demonstration also raises practical questions for deployment at scale: how the device will be maintained underground, how results will be integrated with existing medical records, how clinicians and occupational health teams will be trained to interpret AI‑assisted outputs, and how regulatory oversight will be applied in varied clinical settings. Nexus has described the product as regulated, indicating it has pursued the necessary medical device controls; the report does not detail regulatory approvals or nationwide rollout plans.
Where this fits in South Africa’s health landscape
South Africa faces a high burden of tuberculosis and an ongoing occupational health challenge in the mining sector. Any tool that can triage chest X‑rays rapidly and operate offline could relieve pressure on limited specialist capacity and speed up referral or treatment decisions. For patients, the benefit is practical: faster identification of abnormalities could mean quicker access to confirmatory testing and care.
At the same time, successful integration will depend on funding, procurement choices by public and private occupational health services, and careful evaluation of diagnostic performance in real‑world conditions beyond a single demonstration. The Mponeng test shows that running an AI medical device at the edge is possible; the longer task will be turning that possibility into a reliable, scalable service for workers and communities across South Africa.
