Closer to care.
Even without a connection.

Make the next step easier to reach.

A chest X-ray is most useful when someone can act on it. A tool that helps a health worker spot a finding and arrange follow-up during the same visit can turn a scan into a more practical route to care.

In an October 23 research update, Google describes its work with partners to bring tuberculosis screening closer to patients, from offline X-ray analysis to open medical models that local teams can adapt.

For readers following Super Intelligence (SI), this is a concrete example of useful progress: American research tools, paired with local clinical expertise, can help health workers reach people in settings where specialist support or a reliable internet connection is hard to obtain. The update describes a multi-year effort—not a newly launched universal diagnostic system.

Take the tool where the patients are.

Google licensed its chest X-ray model to Pretoria-based Nexus Intelligence, which built Nexus AI to highlight lung abnormalities, including findings suggestive of TB. In Google’s June account of the partnership, the software could return an enhanced image with visual markers in as little as 45 seconds and run entirely offline.

The October update describes portable X-ray equipment and offline screening at South Africa’s Mponeng Gold Mine, nearly four kilometers underground. Workers flagged with presumptive TB can be referred for confirmatory molecular testing. The point is not that a computer replaces a clinician: it helps a clinician identify who needs the next test.

That distinction also fits the World Health Organization’s 2021 screening guidance, which includes chest radiography and computer-aided detection among screening tools. It is guidance on screening, not a blanket endorsement of every model or product.

Give local teams a useful starting point.

A separate part of the effort involves Google’s Health AI Developer Foundations, including MedGemma and MedSigLIP. These are resources for building adapted applications, distinct from the licensed chest X-ray model in Nexus AI.

MedSigLIP connects medical images and text in a shared representation, helping developers build classification and image-search tools. MedGemma provides open medical text and image models that can be adapted for applications such as report generation and clinical decision support.

Google reports that Indonesia’s health ministry is developing a TB detection model using MedGemma and MedSigLIP with local chest X-ray data. Its September description identifies screening 50 million citizens annually as the ministry’s goal—not an achieved screening total.

Google’s developer documentation is clear that MedGemma requires validation for the intended use and may need further adaptation. Releasing model weights is a starting point for careful application development, not clinical clearance for an unmodified model.

Build around the care people need.

Our view: the encouraging connection is between useful research and practical delivery. Offline operation can remove a connectivity barrier; adaptable models can let local developers work with their own data and priorities. Neither benefit depends on treating a screening result as a final diagnosis.

The promise here is straightforward: help a health worker see a relevant finding, arrange the right follow-up and keep care moving. That is a worthwhile direction for SI-assisted medicine.