Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2019 Oct 18;9(1):15000.
doi: 10.1038/s41598-019-51503-3.

Using artificial intelligence to read chest radiographs for tuberculosis detection: A multi-site evaluation of the diagnostic accuracy of three deep learning systems

Zhi Zhen Qin  1 , Melissa S Sander  2 , Bishwa Rai  3 , Collins N Titahong  2 , Santat Sudrungrot  3 , Sylvain N Laah  2   4 , Lal Mani Adhikari  3 , E Jane Carter  5 , Lekha Puri  1 , Andrew J Codlin  1 , Jacob Creswell  6
Affiliations

Using artificial intelligence to read chest radiographs for tuberculosis detection: A multi-site evaluation of the diagnostic accuracy of three deep learning systems

Zhi Zhen Qin et al. Sci Rep. .

Abstract

Deep learning (DL) neural networks have only recently been employed to interpret chest radiography (CXR) to screen and triage people for pulmonary tuberculosis (TB). No published studies have compared multiple DL systems and populations. We conducted a retrospective evaluation of three DL systems (CAD4TB, Lunit INSIGHT, and qXR) for detecting TB-associated abnormalities in chest radiographs from outpatients in Nepal and Cameroon. All 1196 individuals received a Xpert MTB/RIF assay and a CXR read by two groups of radiologists and the DL systems. Xpert was used as the reference standard. The area under the curve of the three systems was similar: Lunit (0.94, 95% CI: 0.93-0.96), qXR (0.94, 95% CI: 0.92-0.97) and CAD4TB (0.92, 95% CI: 0.90-0.95). When matching the sensitivity of the radiologists, the specificities of the DL systems were significantly higher except for one. Using DL systems to read CXRs could reduce the number of Xpert MTB/RIF tests needed by 66% while maintaining sensitivity at 95% or better. Using a universal cutoff score resulted different performance in each site, highlighting the need to select scores based on the population screened. These DL systems should be considered by TB programs where human resources are constrained, and automated technology is available.

PubMed Disclaimer

Conflict of interest statement

The authors declare no competing interests.

Figures

Figure 1
Figure 1
Frequency distribution of the abnormality scores of CAD4TB, Lunit, and qXR.
Figure 2
Figure 2
The ROC curves of CAD4TB (v6), Lunit (v4.7.2) and qXR (v2) using Xpert results as the reference (n = 1196).
Figure 3
Figure 3
The ROC curves of CAD4TB (v6), Lunit (v4.7.2) and qXR (v2) among individuals with negative smear (n = 1102)
Figure 4
Figure 4
(a) The ROC curves of CAD4TB (v6), Lunit (v4.7.2) and qXR (v2) in Nepal (n = 515). (b) The ROC curves of CAD4TB (v6), Lunit (v4.7.2) and qXR (v2) in Cameroon (n = 681).
See this image and copyright information in PMC

References

    1. Goodfellow, I., Bengio, Y. & Courville, A. Deep learning. (MIT press, 2016).
    1. Krizhevsky, A., Sutskever, I. & Hinton, G. E. In Advances in neural information processing systems. 1097–1105.
    1. Schmidhuber J. Deep learning in neural networks: an overview. Neural networks: the official journal of the International Neural Network Society. 2015;61:85–117. doi: 10.1016/j.neunet.2014.09.003. - DOI - PubMed
    1. World Health Organization. Global Tuberculosis Report 2018. (2018).
    1. Silverman C. An appraisal of the contribution of mass radiography in the discovery of pulmonary tuberculosis. American review of tuberculosis. 1949;60:466–482. - PubMed

Publication types

MeSH terms