Researchers at the University of KwaZulu-Natal, South Africa, together with colleagues from Germany, the UK, and the USA, have devised a computer-aided technique for interpreting digital chest radiography images and identifying new cases of tuberculosis (NPJ Digital Medicine, July 16). People from a rural area of South Africa who had symptoms of tuberculosis or scored above an algorithm’s screening threshold were referred for microbiological sputum assessment, which identifies the bacilli that cause the disease and detects up to 80% of cases. Based on radiographs from 9,914 study participants, the software diagnosed 99 people (1% of the total) with microbiologically proven tuberculosis. The algorithm showed the same sensitivity (percentage of positive results among people with a disease) as assessments made by a radiologist, although with lower specificity (percentage of negative results in people without the investigated disease). A subsequent version of the algorithm was able to match the sensitivity and specificity of the radiologist. Most of the identified cases (78.8%) were asymptomatic.
Republish
Print issue
Current | Past
