TB fight gets shot in arm from AI's diagnosis, Wadhwani AI, Qure.ai lead

Integrating artificial intelligence (AI) in radiology for early detection of tuberculosis (TB) is gradually gathering pace with several Indian health tech companies building deep learning models for screening and analysing abnormalities associated with the infectious disease.

Delhi-based independent non-profit organisation Wadhwani AI and Mumbai-based startup Qure.ai are among those that have started to build AI-based solutions for TB screening and detection.

While Wadhwani AI is working on the development and deployment of AI solutions in partnership with the union health ministry and agencies such as Central TB Division (CTD) and USAID, Qure.ai is looking to leverage deep learning technology to analyse chest radiographs and detect abnormalities associated with TB. 

Studies suggest that AI algorithm-based TB screening can lead to a jump in notifications and help identify 30 to 40 per cent more incidental cases that would have been left undiagnosed in India, which already has a large share of the worldwide burden of the infectious disease.

Experts, however, believe that while the integration of AI models can help in early TB detection due to their ability to analyse medical images with enhanced precision, a permanent solution might still be far away.

Highlighting the potential of AI-based solutions in TB screening, Shibu Vijayan, Chief Medical Officer - Global Health, Qure.ai said that it could become the primary method for TB screening in the next five years as AI can help better predict individual and population-level vulnerability to TB. 

“AI can analyse chest X-rays with high accuracy and identify signs of TB that may be missed by human radiologists. X-rays are very sensitive in detecting lung abnormalities even before symptoms like cough are present”, he said.  

Speaking of their AI tools, Nakul Jain, director, products and programmes, Wadhwani AI said that their AI model employs advanced computer vision to automate the interpretation of Line Probe Assay (LPA) strips for diagnosing drug-resistant TB.

“Our AI tool named ‘Cough Against TB’ uses cough sounds and symptom data collected via a mobile app used by healthcare workers, enabling cost-effective and early TB detection”, Jain said. 

A cough detector AI model validates that the audio recordings contain cough bouts. After successful verification, a cough-based TB classifier (AI model) predicts the TB likelihood score, while a tabular AI model also estimates the TB score based on the symptoms, their duration, and comorbidities. 

“The outputs of these two models are combined into an ensemble AI model to predict whether the person is likely to have pulmonary TB,” Jain added. 
 

Similarly, Qure.ai's qXR software can identify over 30 abnormalities in chest X-rays, providing detailed insights into the presence of TB and other lung diseases.

“qXR is capable of incidental case finding, helping to identify TB in populations undergoing X-rays for other reasons, thereby improving overall detection rates,” Vijayan said. 

The solution has been deployed in over 90 countries and 2700 sites, scanning about a million chest X-rays annually, demonstrating its scalability and effectiveness, he said. 

While AI solutions for screening TB hold tremendous promise, they can face common challenges due to data integration and variability in healthcare settings.

Jain said that some of the challenges from real-world deployment reveal patient refusal, often due to TB stigma, as a significant barrier, especially in convincing asymptomatic individuals to undergo testing. 

“Additionally, in active case-finding settings, individuals may not share their symptoms properly, reducing the effectiveness of symptom-based screening”, he added.
 

Vijayan added that another challenge can be the proliferation of unregulated AI solutions.  “As this is a new area, many AI products are available as open-source, free, or cheap options. However, many do not meet the standards set by the Government of India,” he said.  

Jain adds that such issues are typical for any new technology and are being actively addressed through better frameworks, protocols, and continuous refinement. 

“As these systems evolve and improve, we can expect more seamless integration and enhanced reliability across diverse healthcare environments,” Jain said.  

While the technology is in its early validation stages, if successful, it could help identify individuals for X-ray screening, where AI can then identify probable TB cases for confirmatory testing.