M42 announces results for its AI-powered tuberculosis screening

M42 says it has proof that AI can shoulder the grunt work of tuberculosis screening at scale. In a study published in npj Digital Medicine, the company — working with Capital Health Screening Centre (CHSC) in Abu Dhabi — analyzed more than one million chest X-rays to validate its AI model, AIRIS-TB. The headline numbers are blunt and bright: an AUROC of 98.5% and a reported 0% false-negative rate for TB-specific cases across the full dataset.

As M42's Group CEO Dimitris Moulavasilis put it, "This landmark study marks a pivotal moment in the potential power of AI in the global fight against tuberculosis."

How does it work?

AIRIS-TB is built to triage routine chest X-rays so radiologists can focus on the tricky, urgent, or ambiguous cases. Think of it as a smart bouncer at the clinic door — moving the obvious cases through quickly so specialists can spend time where it matters most.

Key details from the study:

• Scale: Over 1,000,000 real-world CXRs reviewed in collaboration with CHSC.
• Accuracy: AUROC 98.5% for triaging TB on chest X-rays.
• Safety signal: 0% false-negative rate for TB-specific cases across the full dataset.
• Automation potential: Could safely automate up to 80% of routine CXR assessments in high-throughput, low-prevalence settings.
• Generalizability: Strong performance across gender, age, HIV status, income levels, and populations covering six WHO regions.

In short, AIRIS-TB slots into existing workflows as a front-line screener, reducing manual load while flagging what deserves a human eye.

Why does it matter?

Tuberculosis is still a heavy hitter. The WHO estimates 10.8 million people fell ill and 1.25 million died from TB in 2023, with recent reports suggesting TB has returned as the leading cause of death from a single infectious agent. Meanwhile, routine CXR review is tiring, repetitive work — misses rise when speed doubles (one prior study cited 26.6% more missed findings) and errors climb after 9 hours into a shift. That's a recipe for delays and oversight.

If AIRIS-TB performs as reported:

• Fewer misses, faster answers: High sensitivity and triage speed can cut time to action.
• Relief for short-staffed systems: Offloading routine reads helps radiologists tackle complex cases.
• Cost efficiency: Automating the bulk of low-risk screens saves money where volumes are massive.
• Equity: Consistent performance across demographics supports fairer access to early detection.

Dr. Laila Abdel Wareth, CEO of CHSC, didn't mince words: "The outcomes of this study reaffirm that AI models like AIRIS-TB can not only match - but safely surpass - human-level accuracy and efficiency in clinical practice." That's a bold claim — one that, if replicated elsewhere, could shift how high-volume screening gets done.

The context

This isn't a toy demo. M42 frames the project as one of the largest real-world clinical validations of an AI screening tool to date, with peer review in a reputable journal and ethical oversight by the Department of Health - Abu Dhabi. The model's reported stability across varied populations — spanning six WHO regions — suggests robustness beyond a single site or cohort.

It also fits a bigger story: the UAE's push to be a data-driven health-tech hub, and AI's march from research posters to clinic floors. Moulavasilis argues the technology is built for places "where there is a shortage of radiologists and the need to tackle TB is greatest." If subsequent deployments confirm these results in other health systems, AIRIS-TB could become a practical lever to expand capacity, speed up triage, and, yes, save lives — without asking radiologists to sprint for nine hours straight.