Project Details

Evaluating the Accuracy of AI-powered Paediatric TB Diagnostic Pathway: A two-staged study

Heather Zar, MBBCh, PhD

Summary

BACKGROUND: Pulmonary Tuberculosis (PTB) is a major health problem in children globally, with under-diagnosis leading to untreated disease, resulting in substantial mortality and morbidity. Diagnosis is particularly challenging in young children due to paucibacillary disease, non-specific symptoms, and limited microbiological confirmation. Diagnosis of paediatric PTB is highly dependent on chest Xray (CXR) findings. GAP: There is limited expertise in interpreting CXRs in children, with significant inter- and intra-reader variability in identifying key signs of PTB. While WHO recommends AI-assisted CXR interpretation for adults, data on its use in paediatric PTB remain scarce. HYPOTHESIS: • AI-based CXR interpretation can accurately detect abnormalities of PTB and severity in children. • When aided by AI, clinicians will more accurately diagnose PTB in children on CXR compared to routine standard-of-care, unaided by AI. • AI assistance improves accuracy of clinicians in diagnosing paediatric PTB on CXR compared to standard unaided interpretation. METHODS: Children up to 14 years with suspected PTB at three study sites (2 in Pakistan, 1 in South Africa) will be prospectively enrolled. Analysis will be two staged: • STAGE I: Multi-site prospective diagnostic accuracy study using data from 300 confirmed and unconfirmed PTB cases and 250 non-TB LRTI cases using consecutive sampling. CXRs will be blindly analysed by 3 expert radiologists. • STAGE II: Multi-Reader Multi-Case study including 16 clinicians of varying expertise to investigate diagnosis of PTB on 350 CXRs when aided by AI compared to when unaided, using CXRs from stage 1. RESULTS: Pending. IMPACT: An AI tool for paediatric CXRs will augment clinician diagnosis and contribute to improving diagnostic accuracy and treatment for PTB, especially in resource limited settings. The open access database will provide a resource for fostering innovation in global health. Website Link: https://www.gulabdevi.org/pediatric/ https://chich.edu.pk/ https://www.uib.no/en/cih/137682/tuberculosis-research-group https://lunginstitute.co.za/child/ https://health.uct.ac.za/paediatrics-and-child-health https://www.qure.ai/Optional/Additional Comments: This multicentre study spans health facilities across continents, including tertiary care institutions in Pakistan, and South Africa, with academic collaboration from the University of Bergen, Norway, and the University of San Francisco, USA, with AI technical expertise from Qure.ai.