Using AI system to detect active tuberculosis in a high-prevalence setting on CT scans: a multi-center study

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Tuberculosis (TB) remains a global health challenge, particularly in high-prevalence settings.
• Accurate and timely diagnosis of active tuberculosis (ATB) is crucial for effective treatment and control.
• CT imaging is a valuable tool, but interpretation can be challenging, especially in resource-limited or overwhelmed settings.

Purpose of the Study:

• To assess the feasibility and generalizability of an Artificial Intelligence (AI) system for identifying active tuberculosis (ATB) in CT images.
• To evaluate the AI system's performance in TB-specialized hospitals within high-prevalence regions.
• To provide insights for implementing AI to support clinical decision-making in TB diagnosis.

Main Methods:

• Retrospective validation of an AI system using a multi-center dataset of 1741 CT images from three TB-specialized hospitals.
• Dataset included cases of ATB, pneumonia, pulmonary nodules, and normal controls.
• Assessed system utility and generalizability across four application scenarios, with pairwise performance comparisons between hospitals.

Main Results:

• The AI system demonstrated good generalizability across the three participating hospitals.
• Achieved high Area Under the Curve (AUC) values: >0.9 (abnormal vs. normal), >0.95 (ATB vs. normal), >0.8 (ATB vs. non-ATB).
• AUC for distinguishing ATB from other abnormalities (pneumonia, nodules) ranged from 0.762 to 0.906, with no significant performance differences between hospitals in several comparisons.

Conclusions:

• The use of an AI system for identifying ATB in CT images is feasible in TB-specialized hospitals.
• The AI system shows potential for supporting clinical decisions and optimizing resource allocation in TB-burdened healthcare facilities.
• This study offers valuable data for the implementation of AI tools in managing TB cases.