This research aims to evaluate the effectiveness of computer-aided detection (CAD) in classifying disease severity of tuberculosis in children's chest radiographs.
Combined three chest radiograph datasets from children with diagnosed tuberculosis.
CXR interpretations were independently classified by two expert readers as severe or non-severe.
Compared CAD scores generated by CAD4TB and qXR software against human classifications.
Median CAD scores were significantly lower for non-severe versus severe classifications by human readers.
Area under the receiver operating curve was 0.82 and 0.78 for qXR, and 0.79 and 0.76 for CAD4TB respectively.
The difference in CAD scores was greatest in children older than 5 years.