Machine and Deep Learning for Tuberculosis Detection on Chest X-Rays: Systematic Literature Review

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Tuberculosis (TB) is a leading infectious cause of mortality globally.
• Chest radiography (CXR) plays a crucial role in TB detection and diagnosis.
• Human interpretation of CXRs has significant inter-reader variability, impacting reliability.

Purpose of the Study:

• To systematically review and assess the performance of machine learning (ML) and deep learning (DL) algorithms for TB detection using CXR.
• To evaluate the diagnostic accuracy of AI-based methods compared to traditional human interpretation.

Main Methods:

• Systematic literature review (SLR) following PRISMA guidelines.
• Searched Scopus, PubMed, and IEEE databases for relevant studies.
• Included 47 studies, performed risk of bias assessment (QUADAS-2), and conducted meta-analysis on 10 studies.

Main Results:

• Deep learning (DL) was more prevalent (34 studies) than machine learning (ML) (7 studies).
• ML achieved higher mean accuracy (~93.71%) and sensitivity (~92.55%), while DL showed better mean AUC (~92.12%) and specificity (~91.54%).
• Pooled sensitivity and specificity for ML and DL methods were estimated at 0.9857 and 0.9805, respectively.

Conclusions:

• Both ML and DL demonstrate significant potential for TB detection via CXR.
• Future research should prioritize addressing risks of bias, particularly concerning the reference standard and study flow/timing.
• Only two studies developed practical applications from their proposed AI solutions.