Related Experiment Video
Updated: Jun 4, 2026
Three-Dimensional Reconstruction for the Whole Lung with Early Multiple Pulmonary Nodules
Published on: October 13, 2023
Artificial intelligence in paediatric chest imaging: applications, challenges, and future directions
John Joseph Muringathuparambil1, Shamiek Maharaj1, Bradley Max Segal1
1Faculty of Health Sciences, University of the Witwatersrand, 7 York Road, Parktown, Johannesburg, 2193, South Africa.
Insights
Artificial intelligence (AI) in paediatric imaging requires child-specific design, not adaptation of adult systems. Future tools must prioritize safety, equity, and tailored deployment in diverse healthcare settings, especially in low- and middle-income countries.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Healthcare
- Paediatric Radiology
Background:
- Paediatric chest imaging is crucial for diagnosing childhood diseases, particularly pneumonia in low- and middle-income countries (LMICs).
- Radiology expertise is scarce in LMICs, creating a need for AI to improve access and standardize care.
- Current AI systems are often trained on adult data, raising safety and equity concerns for paediatric use.
Purpose of the Study:
- To review AI applications in paediatric chest radiography, lung ultrasound, CT, and MRI.
- To focus on AI integration within LMIC healthcare workflows.
- To identify requirements for safe, paediatric-specific AI deployment.
Main Methods:
- Narrative and pictorial review of published literature.
- Real-world evaluation of a vision-language model (MedGemma) on adult and paediatric chest radiographs.
- Illustration with clinical cases to demonstrate AI performance and failure modes.
Main Results:
- AI demonstrates potential in quality assurance, ultrasound guidance, and multilingual reporting.
- Real-world evaluations revealed critical failures (e.g., pericardial effusion, tuberculosis detection) without clinical context.
- Key challenges include infrastructure, data scarcity, and the need for age-aware, explainable AI models.
Conclusions:
- AI tools for children must be designed specifically for paediatric needs and environments.
- Prioritize federated learning, multimodal integration, and robust validation across age groups.
- Emphasize multilingual communication and caregiver engagement for effective AI deployment.
Background:
Paediatric chest imaging is central to diagnosing respiratory and cardiopulmonary disease, particularly in low- and middle-income countries (LMICs) where pneumonia remains a leading cause of childhood mortality and radiology expertise is scarce. Artificial intelligence (AI) could expand access, standardise quality and support task-shifting in these "diagnostic deserts," yet most systems are trained and validated on adult datasets from high-income settings, and paediatric radiographs form only a small minority of major public training cohorts - raising concerns about safety, generalisability and equity when such models are deployed in children.
Objective:
To synthesise current applications of AI across paediatric chest radiography, lung ultrasound, computed tomography and MRI, with emphasis on LMIC workflows, and to define what is required for safe, paediatric-specific deployment.
Materials And Methods:
Narrative/pictorial review of the published literature, complemented by the authors' real-world evaluation of a generalist vision-language model (MedGemma) on adult and paediatric chest radiographs, illustrated with representative clinical cases.
Results:
Beyond diagnosis, AI shows potential in quality assurance, lung-ultrasound guidance and multilingual reporting. Real-world experience from CAD4Kids and from MedGemma's evaluation - including critical failures in detecting pericardial effusion and tuberculosis without explicit clinical context - illustrates common failure modes and the ethical implications of domain shift. Key challenges include infrastructure constraints, dataset scarcity and the need for age-aware, explainable models.
Conclusion:
Rather than adapting adult systems, future tools must be designed from inception for children and the environments in which they live, prioritising federated learning, multimodal integration, robust validation across age strata and multilingual communication with caregivers.
