source: Google Research: Bringing foundational models to the front lines of TB screening
level: technical
google research and nexus ai have moved tb screening from lab benchmarks to certified clinical tools. the system analyzes chest x-rays for signs of tuberculosis and other lung diseases, returning results in as little as 45 seconds. it is now used in about 40 healthcare facilities across six african countries, with over 25,000 chest radiographs evaluated by more than 90 clinicians. the work was presented at idweek 2026 in washington, d.c.
the tool achieved ce mdr class iib certification and iso 13485 compliance, meeting european medical device standards. a prospective multi-site validation published in nejm ai showed high sensitivity across diverse populations. at the mponeng gold mine in south africa, offline ai and portable x-ray kits screen miners four kilometers underground, enabling same-visit triage and confirmatory testing. nexus ai has also provided over 3,000 free screenings to under-resourced clinics.
to scale beyond a single partner, google open-sourced chest x-ray embeddings and medsiglip, a vision-language model, plus medgemma, a family of open-weight multimodal models with 4b and 27b parameters. medgemma has been downloaded over 10 million times. indonesia's ministry of health is fine-tuning these models on domestic data to screen 50 million citizens annually while keeping patient data within national borders. acoustic screening via hear uses cough sounds from smartphones for areas without imaging.
why it matters: open foundation models let local health systems build tb screening tools that run offline and keep patient data sovereign, cutting diagnosis time from weeks to seconds.
source: Google Research: Bringing foundational models to the front lines of TB screening