Researchers presented four AI tools at the Union World Conference on Lung Health in Copenhagen (18–21 November) that aim to make TB detection and monitoring faster, cheaper and more accessible. The conference noted that TB remains the world’s deadliest infectious disease, causing around 1.25 million deaths in 2024, and highlighted gaps in access to standard diagnostics in many vulnerable communities.
The first approach uses breath analysis, combining "breathomics" and machine learning. Teams from Southern University of Science and Technology and Shenzhen Third People’s Hospital collected breath samples with an AveloMask from about 60 TB patients in South Africa. Liang Fu said the non‑invasive test with machine learning can track recovery and indicate early when a patient is doing well, which could enable safer treatment shortening, improve adherence and reduce costs.
The second approach is Swaasa, a cough‑analysis platform developed with AIIMS, JIPMER and Salcit Technologies. Health workers recorded coughs from more than 350 symptomatic participants using smartphones; the algorithm correctly identified underlying conditions in 94% of cases and correctly predicted respiratory disease risk in 87% of cases. The third tool, from the Wadhwani Institute for AI, mapped vulnerability for active case‑finding under India’s National Tuberculosis Elimination Programme by combining over 20 open‑source datasets with anonymised Ni‑kshay surveillance data; in national testing it achieved 71% accuracy in identifying the top 20% of villages most likely to harbour undetected TB. The fourth is Qure.ai’s qXR, an AI chest X‑ray tool for children from birth to 15 years and the first AI‑enabled chest X‑ray cleared in Europe for that full age range. Experts, including Guy Marks and Ketho Angami, emphasised the extraordinary potential of AI but warned that rigorous testing, strong datasets and staff training are essential, and that reliance on AI alone could be risky in complex cases. Several results are still under peer review and wider validation and implementation will be needed before these tools are used at scale.
Difficult words
- machine learning — Computer methods that learn from data to predict
- algorithm — A set of rules or steps for computing
- adherence — Continuing a treatment as prescribed by health staff
- anonymised — Personal data changed so individuals cannot be identified
- surveillance — Systematic collection and monitoring of health data
- validation — Process of proving a tool works correctly
- implementation — Putting a method or system into practical use
Tip: hover, focus or tap highlighted words in the article to see quick definitions while you read or listen.
Discussion questions
- Which benefits and risks of using AI for TB detection do you think are most important, and why?
- How could anonymised surveillance data improve active case‑finding in vulnerable communities?
- What practical steps would health services need to take before implementing these AI tools widely?
Related articles
Keeping healthy habits during the holidays
The holiday season often disrupts regular health routines. Samantha Harden of Virginia Tech recommends realistic, small changes—like habit stacking, gamifying activities, involving others and travel tips—and a broader view of well‑being so breaks are not failure.
Ecuador teams build tech to fight election disinformation
A revived Hacks Hackers chapter in Ecuador held a February conference and a hackathon to tackle electoral disinformation. Three winning teams — Goddard, VeritasAI and PillMind — received prizes, mentoring and support to develop prototypes.
Mental health conditions raise risk of injury
A large study in Nature Mental Health found people with mental health conditions have higher risks of self-harm, assault and unintentional injuries. The research used population records and had US federal funding and calls for better prevention.
