The substantial case detection gap in the field of child tuberculosis (TB) disease is largely driven by inadequate diagnostic tools and approaches. Chest radiographs (CXRs) remain a key component in the evaluation of children and young adolescents (0–15 years) with presumptive TB, aiding clinicians in making the diagnosis and discriminating children with TB from those with other diseases. Widespread use and optimal interpretation of CXR is hampered by a lack of access to well-trained specialists to interpret images. Artificial intelligence CXR interpretation software, termed computer-aided detection (CAD), is now well developed for adults, yet few products have been evaluated in children. The CXR features of child TB are different from those of adults, and as a result, the performance of these CAD algorithms, largely developed for use in adults, will be suboptimal when used in children. Adapting, or fine-tuning adult CAD algorithms, using CXR images from children with presumptive TB, could allow optimisation of these products for use in children. We, therefore, set out to develop a large image and data repository collected from children evaluated for TB (called Catalysing Artificial Intelligence for Paediatric Tuberculosis Research, CAPTURE) with the purpose of evaluating current CAD products and then working with developers and other partners to optimise CAD algorithms for use in children.

We identified approximately 20 studies, from which potentially up to 11 000 CXRs could be used for the proposed project. CXRs and data were eligible for inclusion in the CAPTURE repository if collected from high-quality child TB diagnostic studies that enrolled children with presumptive TB and if CXRs were obtained as part of the baseline assessment. All lead investigators of these studies are members of the CAPTURE consortium. The images and metadata contributed are centrally collated and the key variable of TB case classification as confirmed, unconfirmed or unlikely TB, using an established consensus case definition, is available. All CXRs included in the CAPTURE repository have a consensus radiological interpretation allocated by a panel of independent expert child TB CXR readers who have classified them as ‘unreadable’, ‘normal’, ‘abnormal typical of TB’ or ‘abnormal not typical of TB’. To determine diagnostic performance of existing CAD products, we will evaluate these against a primary composite clinical reference standard (confirmed TB and unconfirmed TB vs unlikely TB), as well as other secondary microbiological and radiological reference standards. A subset of images will be subsequently allocated to a ‘training set’ and made available to developers, academic groups or other parties to either develop novel paediatric CAD products or fine-tune existing adult ones, which will then be re-evaluated by the CAPTURE team using an image subset (‘validation set’) that is independent of the training set.

The CAPTURE study has been approved by Stellenbosch University Health Research Ethics Committee (N22/09/113), with additional ethics approval or waivers by relevant local authorities obtained by consortium members contributing data if required. The final pooled, harmonised and cleaned dataset, as well as the deidentified, renamed CXR images, is stored on a secure cloud-based server. All analyses of existing CAD products, as well as the paediatric-optimised products, will be published in peer-reviewed publications and shared with other stakeholders like the WHO and donor and procurement organisations to guide policy updates and procurement pathways to ensure widespread uptake.

Diagnosing pulmonary tuberculosis (PTB) in children is challenging owing to paucibacillary disease, non-specific symptoms and signs and challenges in microbiological confirmation. Chest X-ray (CXR) interpretation is fundamental for diagnosis and classifying disease as severe or non-severe. In adults with PTB, there is substantial evidence showing the usefulness of artificial intelligence (AI) in CXR interpretation, but very limited data exist in children.

A prospective two-stage study of children with presumed PTB in three sites (one in South Africa and two in Pakistan) will be conducted. In stage I, eligible children will be enrolled and comprehensively investigated for PTB. A CXR radiological reference standard (RRS) will be established by an expert panel of blinded radiologists. CXRs will be classified into those with findings consistent with PTB or not based on RRS. Cases will be classified as confirmed, unconfirmed or unlikely PTB according to National Institutes of Health definitions. Data from 300 confirmed and unconfirmed PTB cases and 250 unlikely PTB cases will be collected. An AI-CXR algorithm (qXR) will be used to process CXRs. The primary endpoint will be sensitivity and specificity of AI to detect confirmed and unconfirmed PTB cases (composite reference standard); a secondary endpoint will be evaluated for confirmed PTB cases (microbiological reference standard). In stage II, a multi-reader multi-case study using a cross-over design will be conducted with 16 readers and 350 CXRs to assess the usefulness of AI-assisted CXR interpretation for readers (clinicians and radiologists). The primary endpoint will be the difference in the area under the receiver operating characteristic curve of readers with and without AI assistance in correctly classifying CXRs as per RRS.

The study has been approved by a local institutional ethics committee at each site. Results will be published in academic journals and presented at conferences. Data will be made available as an open-source database.

PACTR202502517486411