Diagnostic Performance of a Computer-aided System for Tuberculosis Screening in Two Philippine Cities

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Tuberculosis diagnostics

Background:

• The Philippines faces a significant shortage of skilled health workers for tuberculosis (TB) screening.
• Deep learning neural networks (DLNNs) show promise for TB screening using chest radiographs (CXRs).
• Limited local studies exist on AI-based TB screening in the Philippines.

Purpose of the Study:

• To evaluate the diagnostic performance of qXR3.0 technology for TB screening in Filipino adults.
• To assess the sensitivity and specificity of qXR3.0 compared to radiologist interpretations.
• To determine if qXR3.0 meets World Health Organization (WHO) standards for TB screening.

Main Methods:

• Prospective cohort study adhering to STARD guidelines.
• Comparison of qXR3.0 and radiologist readings against the Xpert MTB/RiF assay reference standard.
• Inclusion of 82 adult participants from two Metro Manila clinics.

Main Results:

• qXR3.0 achieved 100% sensitivity and 72.7% specificity.
• Strong agreement observed between qXR3.0 and radiologist readings (concordance indices ranging from 0.7895 to 0.9403).
• The AI tool demonstrated high accuracy in identifying TB presence.

Conclusions:

• qXR3.0 exhibits high sensitivity for TB detection and meets the WHO's 70% specificity standard.
• The technology has significant potential to address the radiologist shortage for TB screening in the Philippines.
• Further research with larger sample sizes and economic evaluations is recommended for widespread adoption.