What happened
Computer-aided detection software reads a digital chest X-ray and returns a score for the likelihood of tuberculosis. WHO says that such systems are trained on thousands or millions of images using artificial intelligence. In 2024, WHO invited makers to submit products for an independent evaluation on a validation platform run by FIND in Geneva. Eight products met regulatory requirements. They were tested on two image libraries, 1,663 images for community screening and 2,252 for facility screening, which the makers had not used for training. Microbiological test results served as the reference, and radiologists read the same images for comparison. To pass, a product needed at least 40% specificity at 90% sensitivity and a high partial area under the curve in both libraries. WHO's Technical Advisory Group reviewed the results in January 2025. In a policy statement published on 10 June 2025, WHO found that six products met its standards. In community screening their specificity at 90% sensitivity ranged from 45% to 56%. Accuracy was generally lower for people living with HIV, women and people with a history of TB.
Sources
-
Assessment report, by authors with no tie to the system.
World Health Organization, Global Programme on Tuberculosis & Lung Health. Use of computer-aided detection software for tuberculosis screening: WHO policy statement. Geneva, 10 June 2025. ISBN 978-92-4-011037-3.
CAD4TB v7, DrAid for TB screening v2.4.4–6, Genki Edge v3.4.2, InferRead DR Chest v1.0.1.1, Insight CXR v3.1 and qXR v4.0 met the WHO performance standards for the use of CAD in community-based and facility-based screening for pulmonary TB in people aged 15 years and older
Details
AI system
- Reported configuration
-
CAD4TB v7, DrAid for TB screening v2.4.4–6, Genki Edge v3.4.2, InferRead DR Chest v1.0.1.1, Insight CXR v3.1 and qXR v4.0
- Workflow
-
Each product was set to a threshold giving 90% sensitivity against microbiological results. The image libraries came from earlier screening studies and were never shared with the makers.
Follow-up
- Conditions for replication
-
Programs calibrate the threshold for their own setting, hardware and population. Programs have laboratory capacity for the confirmatory tests that positive screens require.
Entry details
- Sector
-
Healthcare (tuberculosis screening).
- Purpose
-
Delft Imaging, VinBrain, DeepTek, Infervision, Lunit and Qure.ai built their programs to read digital chest X-rays and score the likelihood of tuberculosis so that screening programs can decide who needs further testing.
Not yet checked: model, provider, role, configuration, inputs and outputs, autonomy, human oversight, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). Six AI chest X-ray readers meet WHO's standard for TB screening (AIBD-0733). Artificial Intelligence Benefits Database, edition of 31 October 2026. https://petervartanian.xyz/aibd/records/AIBD-0733.html (archived at https://doi.org/10.7910/DVN/AADSII, version 4.0)