Google Research details how MedGemma and MedSigLIP bring TB screening to the front lines
Bringing foundational models to the front lines of TB screening
Google Research described how its chest X-ray TB screening work moved from a 2019 Radiology benchmark to a CE MDR class IIb certified tool deployed by partner Nexus AI in about 40 facilities across six countries. Over 90 clinicians have evaluated more than 25,000 chest radiographs, cutting turnaround from weeks to under 45 seconds, and the tool runs offline four kilometers underground at South Africa's Mponeng Gold Mine. Google also open-sourced chest X-ray embeddings and MedSigLIP through HAI-DEF and released MedGemma, a 4B and 27B open-weight multimodal family with more than 10 million downloads, which Indonesia's Ministry of Health is fine-tuning on domestic radiographs for a national program targeting 50 million screenings annually.
The piece traces one TB screening model from benchmark to CE-certified clinic deployment, showing what the translation gap costs.
Source: Google Research · research.google