Start4Kids: improving how chest X-ray is used to diagnose tuberculosis in children

Tuberculosis remains the world's leading cause of death from infection, responsible for more than 1.3 million deaths each year, and children are disproportionately affected. 

Chest X-rays are an important tool for making a TB diagnosis. However, the use of chest X-rays is more complicated in children than in adults, as the radiological signs of TB present differently. 

In many settings where the burden of TB is highest, expertise to interpret these X-rays is often unavailable. Computer-aided detection (CAD), also known as AI-enabled detection, uses software to analyse the chest X-ray image and indicate the risk of TB being present.  

Current World Health Organization guidance supports the use of CAD/AI in patients older than 14, but not in younger children. This reflects the limited evidence currently available to know that it also works reliably in younger children. 

To establish whether software can reliably interpret a child's chest X-ray requires large sets of paediatric chest X-rays that have been read by experienced radiologists. Such datasets exist, but more are still needed. They can also help train and improve AI models that are under development.

Our work in Start4All: Phase 1

Start4All is research initiative launched in 2022, funded by UNITAID and co-led by the Liverpool School of Tropical Medicine and the Stop TB Partnership. The first phase included seven countries - Bangladesh, Brazil, Cameroon, Kenya, Malawi, Nigeria and Vietnam - with teams researching gaps in the accessibility, affordability and equity of TB diagnosis for adults and children.

Worldwide Radiology joined the collaboration in early 2024, to support the development of a library of 1,000 children’s chest X-rays of sufficient quality to train and validate software for childhood TB detection. 

This involved technical input into how images were best processed, quality assurance of the X-rays and reads. We worked alongside AI software specialists to ensure the dataset was of high quality. Our team also contributed technical expertise to study teams for safety and general use of chest X-ray in TB in children.

"The work we did in Start4All demonstrated the importance of developing high quality chest X-ray datasets to improve computer aided detection of TB in children. In phase 2 of the project, Start4Kids, we will use our experience and learnings to develop an even bigger and better chest X-ray library for CAD evaluation and training, potentially increasing the impact in diagnosing TB in this vulnerable group." - Laura Chillag, Project Lead at Worldwide Radiology

Start4Kids: a Phase 2 follow-up study

For the Start4Kids study, we will be working with the same teams at LSTM and partner sites, supporting all Chest X-ray related work, while they will continue to gather more information on new laboratory tests for TB.

This study will be on a much larger scale, enrolling up to 4,300 children aged 0 to 14 across six sites in Kenya, Bangladesh, Cameroon and Vietnam. 

Each X-ray will be read independently by a team of 12 radiologists. We will improve how we capture findings that are not due to TB, because AI developers are now designing software that may detect other diseases.

The study sites will include hospitals with X-ray departments, but also primary health centres in Kenya where ultra-portable X-ray machines will be used to take children’s X-rays, with regional radiologists providing remote oversight. This allows the study teams to examine what is required for children to safely and effectively access Chest X-ray tests at the community healthcare level, where most children with suspected TB first present. It is at these levels that the need for better access is greatest.

Worldwide Radiology’s contribution

Worldwide Radiology provides the paediatric radiology and radiography expertise across the study's imaging components:

• Input into the study protocol and the reading methodology

• Updating of structured reading proforma, including what Start4All taught us about non-TB findings

• Recruiting, training and supporting the twelve study radiologists, with quality assurance running throughout the reading period

• Providing X-ray technical quality assurance and trouble shooting. 

• Supporting implementation research on ultra-portable X-rays at the primary health centre, exploring technical aspects, patient and staff experience, challenges and opportunities for scale up.  

• Contributing to data analysis, publication and dissemination of results.