Performance of Computer-Aided Detection Software in Tuberculosis Case Finding in Township Health Centers in China

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Public Health

Background:

• Digital chest X-rays are increasingly interpreted using computer-aided detection (CAD) software.
• Evaluating the efficacy of domestic CAD solutions for tuberculosis (TB) case finding in China is crucial.

Purpose of the Study:

• To assess the performance of JF CXR-1 v3.0 CAD software in identifying active tuberculosis cases.
• To compare the diagnostic accuracy of the CAD software with that of local radiologists in a real-world setting.

Main Methods:

• An internal evaluation used expert-annotated standard images.
• A prospective on-site study compared JF CXR-1 v3.0 with local radiologists using chest X-rays from 3705 outpatients.
• Expert panel readings served as the reference standard for TB diagnosis.

Main Results:

• Internal assessment showed JF CXR-1 v3.0 sensitivity of 85.94% and specificity of 74.65% for active TB.
• On-site, JF CXR-1 v3.0 achieved 92.11% sensitivity and 94.54% specificity.
• Local radiologists had 32.89% sensitivity and 99.28% specificity for active TB.

Conclusions:

• CAD software, specifically JF CXR-1 v3.0, demonstrated superior sensitivity in TB case finding compared to local radiologists.
• The findings suggest that CAD software can enhance TB detection capabilities in primary healthcare settings like township health centers.