AI tools poised to transform global TB detection

[SciDev.Net] Researchers have unveiled new AI tools, from smartphone cough analysis to child-friendly screening systems, which could transform how tuberculosis is detected, monitored and prevented.

The innovations, presented at the Union World Conference on Lung Health in Copenhagen, Denmark (18-21 November), show promise for faster, more accessible, personalised TB care, especially for communities where traditional diagnostics remain out of reach.

Guy Marks, president of the International Union against Tuberculosis and Lung Disease, said the advances showed “the extraordinary potential of artificial intelligence to transform the fight against TB and lung disease”.

“The challenge now is ensuring these innovations reach the people and health systems that need them most,” he said at the conference.

TB remains the world‘s most deadly infectious disease, causing around 1.25 million deaths in 2024, according to the World Health Organization (WHO). Many vulnerable communities are hard to reach, underscoring the need for accurate, accessible diagnostics.

Breath test

At the conference, scientists from Southern University of Science and Technology and Shenzhen Third People’s Hospital in China showcased an AI-powered breath analysis system for tracking how patients respond to TB treatment.

“The challenge now is ensuring these innovations reach the people and health systems that need them most.”

Guy Marks, president of the International Union against Tuberculosis and Lung Disease

Researchers evaluated whether so-called “breathomics”—the analysis of chemical compounds in exhaled breath—combined with machine learning could track TB treatment progress more effectively than current methods like sputum culture (laboratory testing of mucus produced by the lungs) or X-ray.

Researchers used their AveloMask to collect breath samples from around 60 TB patients in South Africa and studied subtle chemical changes in their exhaled air with the help of machine learning.

Liang Fu, a pulmonologist and TB specialist at Shenzhen Third People’s Hospital, told the conference: “Our study suggests that a non-invasive breath test combined with machine learning can track recovery during TB treatment and indicate early when a patient is doing well.

“This approach could enable safer treatment shortening, improve adherence, and reduce costs for patients and TB programmes.”

Cough analysis

Researchers from the All India Institute of Medical Sciences (AIIMS), Jawaharlal Institute of Postgraduate Medical Education and Research, and Salcit Technologies presented results from Swaasa, an AI platform that uses a smartphone to analyse cough sounds.

The AI is trained to distinguish coughs caused by TB from those linked to other respiratory illnesses, offering a low-cost diagnostic alternative for settings where X-rays and molecular tests are scarce.

Using a smartphone, health care professionals recorded the coughs of more than 350 participants with cough symptoms. Results given by the AI model were compared to standard test results to find out how accurate it was at identifying respiratory disorders such as TB.

The AI algorithm correctly identified underlying conditions in 94 per cent of cases and correctly predicted the risk of respiratory diseases in 87 per cent of cases, according to the study.

Rakesh Kumar, associate professor at the Centre for Community Medicine, at AIIMS, said the system was suitable for large-scale deployment.

He told the conference: “Screening becomes faster, simpler, and more inclusive, helping fill critical gaps in TB case detection in resource-limited settings.”

Vulnerability mapping

To strengthen active TB case-finding under India’s National Tuberculosis Elimination Programme, researchers from the Wadhwani Institute for AI presented an AI-driven vulnerability mapping system to pinpoint communities most likely to have undiagnosed TB cases.

The system combined over 20 open-source datasets—demographic, geographic, and economic—with anonymised case data from India’s Ni-kshay TB surveillance system.

Aparna Chaudhary, national TB lead at Wadhwani AI, said in national testing this achieved 71 per cent accuracy in identifying the top 20 per cent of villages most likely to harbour undetected TB. The results of the study are currently undergoing peer review.

He said the system aimed to increase precision and efficiency in TB case finding and guide where resources are needed most.

Child screening

Meanwhile, Mumbai-based health tech company Qure.ai announced an AI-powered child TB screening tool, qXR, for use from birth to 15 years. It is the first AI-enabled chest X-ray tool to receive European regulatory clearance for children of this age range, including the youngest.

“The youngest children have long been the hardest to reach and the most vulnerable,” said Shibu Vijayan, chief medical officer, global health at Qure.ai.

“With this tool, we are proud to equip healthcare systems worldwide with a scalable, reliable way to detect TB early, prioritise care, and ultimately save lives.”

Ketho Angami, president of India’s Access to Rights and Knowledge (ARK) Foundation, says new technologies to improve TB detection, diagnosis, and treatment are important, but stresses they must be tested rigorously.

“We need more and better tools,” he told SciDev.Net. “However, the key issue is whether these AI technologies can provide clear and reliable conclusions […].

“If their accuracy, efficacy, and specificity are well validated, then that is a positive development. But without a strong and well-supported dataset, relying on such tools becomes risky.”

People managing AI systems must be trained to interpret the results and “go beyond what the system outputs”, Angami added.

“If AI is fed clear and appropriate parameters about TB, it can offer a clear answer. But if the situation is more complex or unclear, it could become dangerous to depend on it alone,” he warned.

This piece was produced by SciDev.Net’s global desk.