Evaluating InferVision's Computer-Aided Detection (CAD) algorithm for Tuberculosis (TB) screening, Lusaka, Zambia.

Area of Science:

• Radiology and Imaging
• Infectious Disease Diagnostics
• Artificial Intelligence in Healthcare

Background:

• Tuberculosis (TB) remains a significant global health challenge, particularly in high HIV burden regions.
• Accurate and efficient TB screening tools are crucial for early detection and treatment.
• Artificial intelligence (AI) offers potential for improving chest X-ray interpretation in TB screening.

Purpose of the Study:

• To evaluate the diagnostic performance of the InferRead DR Chest AI tool for tuberculosis screening.
• To assess InferRead DR Chest's accuracy in a high HIV and TB prevalence setting in Lusaka, Zambia.
• To determine if the AI tool meets World Health Organization (WHO) target product profiles for TB screening.

Main Methods:

• Retrospective analysis of 1,890 anonymized chest X-ray images from individuals aged 15+.
• Comparison of InferRead DR Chest classifications against a composite reference standard (Xpert MTB/RIF or MTB culture).
• Evaluation of performance using Area Under the Receiver Operating Characteristic Curve (AUC) and sensitivity/specificity at a defined threshold.

Main Results:

• The overall AUC for InferRead DR Chest was 0.81.
• AUCs were 0.71 for individuals with prior TB history and 0.77 for HIV-positive individuals.
• At 90.3% sensitivity, the AI tool achieved 39.2% specificity, with a TB score cut-off of 0.12.

Conclusions:

• InferRead DR Chest demonstrated acceptable diagnostic performance for TB screening in the studied high-burden population.
• The AI tool's performance, particularly specificity, may require optimization for this demographic.
• Recommendations include further training and piloting of InferRead DR Chest in similar settings to enhance its utility.