Tuberculosis (TB) continues to disproportionately decimate people in resource-constrained countries, despite the disease being curable and preventable. Despite several TB containment measures, such as screening and diagnosis, thousands of mortalities are still recorded daily, worldwide, making it a serious global public health concern. The inherent challenges faced by healthcare systems in resource-constrained countries make their residents susceptible to TB-related deaths, especially when coupled with the double burden of disease arising from human immunodeficiency virus (HIV) co-infection. While researchers are increasingly interested in applying artificial intelligence (AI) to TB detection, there is a dearth of reviews that organise such literature. Preferred Reporting Items for Systematic and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) methodology guided the reporting of this scoping review on deep and machine learning models deployed to tackle TB in resource-limited settings. Machine learning algorithms, like logistic regression, k-nearest neighbour, and support vector machine, and deep learning algorithms like long short-term memory and convolutional neural networks, have been used to detect TB and performed well based on the performance evaluation metrics. However, various challenges still exist in resource-constrained countries, such as a shortage of localised, large-scale, real-time, and well-annotated datasets, high chest x-ray hardware expenses, limited integration with local health information systems, limited application in different subgroups, limited capability in detecting non-TB abnormalities, and limited sputum-based detection. Data-driven AI-based tools present unmatched opportunities to end TB in resource-constrained contexts through enhanced, accurate, and timely TB detection. However, concerted efforts are needed to overcome the identified challenges by investing in localised datasets and integrating AI with existing healthcare infrastructure. • Tuberculosis (TB) continues to kill many people globally despite it being curable. • Resource-constrained settings face various challenges in dealing with TB, such as a lack of specialists. • Machine learning can address skills gaps and facilitate timely TB detection, saving lives through timely treatment. • This paper provides insights into the potential role of ML in TB detection in resource-constrained settings. • This lack of specialists often leads to delayed detection and false positives and negatives. • It also provides recommendations on how such technologies can be incorporated into daily clinical practice.
Batani et al. (Sun,) studied this question.