Bringing AI Tools to Tuberculosis Screening Around the World
Google Research shared how artificial intelligence tools and open foundation models are being used to screen for tuberculosis in under-resourced areas. The evidence comes from real-world clinical deployments, prospective validations, and a presentation shared at an event.
In short: Google Research shared how artificial intelligence tools and open foundation models are being used to screen for tuberculosis in under-resourced areas. The evidence comes from real-world clinical deployments, prospective validations, and a presentation shared at an event.
Tuberculosis remains a deadly global disease, but getting a timely chest scan often depends entirely on where a person lives. Google Research and its partners are working to change that by bringing artificial intelligence tools directly to the front lines of healthcare.
What happened, in plain words
Google Research and Google for Health announced updates on their work to use artificial intelligence for tuberculosis screening at an event called IDWeek 2026. Over the past several years, the team has moved from early research benchmarks to clinical trials, regulatory clearances, and open foundation models. Partnering with companies and public health organizations like Nexus AI, they have translated chest X-ray algorithms into certified clinical tools. These tools are now active in about 40 healthcare facilities across countries such as South Africa, Sierra Leone, and Côte d'Ivoire. The technology is also being used deep underground at the Mponeng Gold Mine in South Africa, where portable X-ray kits and offline AI help screen miners quickly. Additionally, Google released open-source models like MedGemma and MedSigLIP, and developed an acoustic screening tool called HeAR to listen for signs of disease in cough sounds recorded on smartphones.
Key points
- Moving from lab to clinics: Algorithms developed by Google Research were turned into a certified clinical diagnostic tool by Nexus AI. This technology is currently operating in about 40 healthcare facilities across six countries.
- Screening deep underground: At the Mponeng Gold Mine in South Africa, ultra-portable digital X-ray kits and offline artificial intelligence allow miners to be screened on-site in seconds instead of waiting weeks.
- Open foundation models: Google released open models like MedGemma and MedSigLIP, which let local developers and health ministries fine-tune software on regional data without needing massive computing resources or exposing private data.
- Acoustic smartphone screening: Google Research created Health Acoustic Representations, or HeAR, to perform non-invasive screening by analyzing cough sounds recorded on standard smartphone microphones.
Terms explained
- Foundation model — A large artificial intelligence model trained on a massive amount of data that can be easily adapted to perform many different specific tasks. Example: Think of it as a very smart, multi-talented student who has read thousands of books and can quickly learn to specialize in medicine, law, or history.
- Chest radiograph — A medical imaging scan of the chest area, commonly known as an X-ray. Example: It is like a black-and-white shadow picture that lets doctors see the inside of your lungs and heart.
- Multimodal model — An artificial intelligence system that can process and combine different types of information, such as both images and written text. Example: This is similar to a human expert who can look at an X-ray picture while simultaneously reading a patient's written medical history to understand the whole story.
Critical analysis
What is encouraging. The work demonstrates a clear pathway from initial deep learning research publications to real-world deployments in underserved regions, addressing a major global health bottleneck by cutting down screening times and utilizing offline capabilities where internet access is limited.
What deserves caution:
- Company-reported achievements. The announcement and statistics come directly from the creators at Google Research, meaning the claims about deployment success and clinical impact are shared by the organization that built the technology rather than an independent audit.
- Varying deployment stages. While some tools are operating in clinics, other components like acoustic smartphone screening are still being explored through collaborative research and grants, meaning practical readiness varies widely across different tools.
- Localization and adaptation hurdles. Although open models allow local fine-tuning, the source does not detail the exact diagnostic accuracy rates across every single new regional dataset or how well the software performs when translated to entirely new healthcare environments.
Bottom line: Google Research is successfully deploying AI screening tools in real-world clinics, but independent verification across all new open-source applications is still developing.
What we still don't know
The source does not provide comprehensive clinical trial data for every single country mentioned, nor does it detail the exact financial costs or long-term health outcomes for patients screened using the new models.
Based on reporting from Google Research. This is an independent explainer, written in our own words with AI assistance; Google Research has not reviewed or endorsed it. Read the original for the full details.