Illustration by tuput
English
The Delhi State Health Mission, the Delhi State TB Elimination Programme and The Union used AI software to pick out X-rays for a sputum test. Of the 18,377 people flagged, 91.7% completed molecular testing, and one confirmed case turned up for every 192 people screened.
Between July 2025 and March 2026, health teams in Delhi X-rayed 1,68,568 residents (about 1.69 lakh) aged 15 and over at 3,566 community camps in 349 urban slum clusters. Software read the images, 18,377 people were flagged, and molecular tests confirmed tuberculosis in 877 of them. That is 520 confirmed cases for every 1 lakh (100,000) people screened, or one case for every 192 screened, according to an evaluation in the Indian Journal of Tuberculosis.
The programme was run by the Delhi State Health Mission, the Delhi State TB Elimination Programme and the International Union Against Tuberculosis and Lung Disease, India (The Union). The paper went online on 23 August 2026 and appears in the journal’s September 2026 supplement, according to the Europe PMC and Crossref records. PTI reported its findings on 6 October 2026.
Where the camps were and who came
The 349 clusters are notified urban slums in six Delhi districts, with an estimated population of 18.3 lakh, PTI reports from the paper. Screening was scaled up in July 2025 and the evaluation was carried out in April 2026. Every resident aged 15 or above in the chosen communities was eligible, whether or not they had symptoms or known risk factors.
The camps ran at Anganwadi centres (government child and mother care centres), schools, community halls and private health facilities. The paper’s abstract says the programme used ultra-portable chest X-ray devices.
What the software did, and what it did not
The AI software was Qure.ai’s, according to PTI. It read digital chest X-rays taken at the camps and marked the images that looked suggestive of TB. It did not diagnose anyone. People whose X-rays were flagged were offered a sputum collection, and the samples travelled through the programme’s existing specimen network to a molecular test, which is what confirmed or ruled out the disease.
The software decided who should give a sample first. The laboratory decided who had TB. Of the 1,68,568 people X-rayed, 18,377 (10.9%) were flagged. Of those, 16,866 (91.7%) completed molecular testing. The test confirmed TB in 877 people, which is 5.2% of those tested and 0.52% of everyone screened.
The researchers say that bringing sputum collection to the screening site, and using the existing transport network for samples, helped flagged people finish testing. They also credit ASHAs (accredited social health activists, the community health workers) and screening at familiar local venues for participation.
Why screen with an X-ray at all
Symptom checks miss people. The authors cite the National TB Prevalence Survey, in which more than 40% of the bacteriologically confirmed cases found did not report the symptoms normally used to pick people for testing, PTI reports. The researchers say X-ray screening with an AI reader could address that limitation. The paper’s abstract adds that equipment downtime was low, at 3.5% of planned operating days.
What WHO and ICMR say about this kind of tool
WHO’s screening guidelines, published on 22 March 2021, list computer-aided detection software, which reads a digital chest X-ray, among the recommended screening tools. On 11 June 2025, WHO said six such products met its performance standards for screening people aged 15 and over, after an independent evaluation that began with an assessment by the Foundation for Innovative New Diagnostics in Geneva. WHO does not yet recommend the software for children and adolescents under 15, and says people who screen positive should have confirmatory testing before starting treatment. The pages tuput opened do not say which products passed, so this article does not say whether Qure.ai’s did.
India also has its own tool. The Indian Council of Medical Research (ICMR), its Department of Health Research and the Institute for Plasma Research built DeepCxR version 1.1, which sorts chest X-rays into normal and abnormal and can identify findings including pulmonary TB. ICMR says the tool will be distributed free under the National TB Elimination Programme, can process over 250 images in 30 minutes, and that flagged cases can then be reviewed and confirmed by qualified medical experts.
Minister of State for Health Anupriya Patel told the Rajya Sabha on 28 March 2025, in a written reply reported by PTI, that the National TB Elimination Programme had adopted DeepCXR, that ICMR had validated and recommended the AI tools in use, and that the tool was deployed in eight states and Union Territories. The same reply said handheld X-ray machines had been approved for central procurement for high-burden areas. In a Lok Sabha reply reported by IANS on 6 December 2025, Minister of State Prataprao Jadhav listed an Abnormal Chest X-ray Classifier Model among the health ministry’s AI solutions. The government health cover, Ayushman Bharat PM-JAY, is a separate scheme with its own article.
The sources opened for this article do not say whether Qure.ai’s software is part of the national programme, and the ICMR page gives no validation figures for DeepCxR. PTI names Qure.ai, not DeepCxR, as the software used in Delhi.
What the study does not show
This article rests on the journal’s abstract record and two news reports of the full paper, because the publisher’s own page did not load for tuput. The paper reports what the programme did and found. Its statistics are descriptive, and the abstract and the news reports mention no comparison group and give no sensitivity or specificity figures. That means they cannot say how many people with TB the software failed to flag, or how many a different method would have found. A randomised comparison would answer that, and the Kenya trial of an AI assistant for clinicians shows what one looks like for a different health tool. Three further limits come from the paper’s own numbers.
- Reach. An independent household survey in the clusters estimated that 30.5% of residents had been screened (95% confidence interval 24.3% to 36.8%). The authors name low perceived need among people without symptoms and competing work as the main barriers, and say ASHAs and doorstep services helped uptake.
- Follow-up. By subtraction, 1,511 of the 18,377 flagged people (about 8%) did not complete molecular testing, and the figures do not include their results.
- Flags that did not end in a TB diagnosis. The software flagged 10.9% of images, and TB was confirmed in 5.2% of the flagged people who were tested. The sources do not say what the other flagged X-rays showed.
The authors conclude that community-wide AI-assisted chest X-ray screening was feasible at scale and reached a high completion rate across the diagnostic steps, and that stronger community participation could raise its public health impact.
Sources & further reading
- Europe PMC: Evaluating community-wide AI-assisted systematic tuberculosis screening in urban slums of Delhi (Saini and others, Indian Journal of Tuberculosis, vol. 73, supplement 2, pages S190 to S194, first published 23 August 2026)
- Crossref: record for DOI 10.1016/j.ijtb.2026.08.015, the Delhi evaluation (created 23 August 2026, print issue September 2026)
- ThePrint, carrying a PTI report: AI-assisted X-rays flag 18,377 people for TB testing in Delhi slums, 877 confirmed (6 October 2026)
- The Pioneer: AI chest x-ray screening flags 18,000, confirms 877 TB cases in slum clusters (8 October 2026)
- ICMR: About AI4CXR, the DeepCxR version 1.1 chest X-ray screening tool
- Business Standard, carrying a PTI report: AI tools used in TB screening under 100-day drive, Govt to Rajya Sabha (28 March 2025)
- IANS: AI-based diagnostic tools boosting TB, diabetes care, Govt (6 December 2025)
- WHO: Consolidated guidelines on tuberculosis, Module 2: screening, systematic screening for tuberculosis disease (22 March 2021)
- WHO: WHO approves six software products for computer-aided detection of TB on chest X-ray (11 June 2025)
- WHO: Use of computer-aided detection software for tuberculosis screening, WHO policy statement (June 2025)
Researched and written with the help of AI tools and edited for accuracy. Provided for general information and discussion only, not professional advice. See our editorial standards and disclaimer. Spotted an error? Tell us.
Enjoyed this? Get the next one.
One good read at a time, straight to your inbox. No spam, unsubscribe anytime.