Publication date: Jul 25, 2025

Artificial intelligence (AI) has become an important tool in global health, improving disease diagnosis and management. Despite advancements, tuberculosis (TB) remains a public health challenge, particularly in low- and middle-income countries where diagnostic methods are limited. In this scoping review, we aim to examine the potential role of AI in TB control. We conducted a search on 25 August 2024 for the past five years, in the PubMed database using keywords related to AI and TB. We included laboratory-based and observational studies focussing on AI applications in TB, excluding non-original research. There were 34 eligible studies, identifying eight overarching aspects associated with TB control, including active case finding (ACF), triage, pleural effusion diagnosis, multidrug-resistant (MDR) TB and extensively drug-resistant (XDR) TB, differential diagnosis distinguishing active TB from TB infection and other pulmonary communicable diseases, TB and other pulmonary communicable and non-communicable diseases (NCDs), treatment outcome prediction, pleural effusion, and predictions of regional and national trends. AI may transform TB control through enhanced ACF methods and triage, improving detection rates in high-burden regions. With high accuracy, AI may diagnose pleural diagnosis, differentiate TB active and TB infection, TB and non-tuberculous mycobacterial lung disease, COVID-19, and pulmonary NCDs. AI applications may facilitate the prediction of treatment success and adverse effects. Furthermore, AI-driven hotspot mapping may identify undiagnosed TB cases at rates surpassing traditional notification methods. Lastly, predictive modelling and clinical decision support systems may improve the management of MDR-TB. This scoping review highlights the potential of AI-driven predictions in national TB programmes to enhance diagnostics, track trends, and strengthen public health surveillance. While promising for reducing transmission and supporting TB care in low-resource settings, these models require large-scale validation to ensure real-world applicability, especially for high-risk groups.

Semantics

Original Article