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
. 2018 May 1;22(5):567-571.
doi: 10.5588/ijtld.17.0492.

Accuracy of an automated system for tuberculosis detection on chest radiographs in high-risk screening

J Melendez  1 , L Hogeweg  2 , C I Sánchez  2 , R H H M Philipsen  3 , R W Aldridge  4 , A C Hayward  5 , I Abubakar  6 , B van Ginneken  3 , A Story  4
Affiliations

Accuracy of an automated system for tuberculosis detection on chest radiographs in high-risk screening

J Melendez et al. Int J Tuberc Lung Dis. .

Abstract
in English, French, Spanish

Setting: Tuberculosis (TB) screening programmes can be optimised by reducing the number of chest radiographs (CXRs) requiring interpretation by human experts.

Objective: To evaluate the performance of computerised detection software in triaging CXRs in a high-throughput digital mobile TB screening programme.

Design: A retrospective evaluation of the software was performed on a database of 38 961 postero-anterior CXRs from unique individuals seen between 2005 and 2010, 87 of whom were diagnosed with TB. The software generated a TB likelihood score for each CXR. This score was compared with a reference standard for notified active pulmonary TB using receiver operating characteristic (ROC) curve and localisation ROC (LROC) curve analyses.

Results: On ROC curve analysis, software specificity was 55.71% (95%CI 55.21-56.20) and negative predictive value was 99.98% (95%CI 99.95-99.99), at a sensitivity of 95%. The area under the ROC curve was 0.90 (95%CI 0.86-0.93). Results of the LROC curve analysis were similar.

Conclusion: The software could identify more than half of the normal images in a TB screening setting while maintaining high sensitivity, and may therefore be used for triage.

CONTEXTE:: Les programmes de dépistage de la tuberculose (TB) peuvent être optimisés en diminuant le nombre de radiographies pulmonaires (CXR) qui requièrent une interprétation par des experts.

OBJECTIF:: Evaluer la performance de logiciels de détection numérique pour le tri des CXR au sein d'un programme de dépistage de TB mobile numérique à haut débit.

SCHÉMA:: Une évaluation rétrospective du logiciel a été réalisée sur une base de données de 38 961 CXR de face réalisées entre 2005 et 2010 chez des patients dont 87 ont eu un diagnostic de TB. Le logiciel a généré un score de probabilité de TB pour chaque CXR. Ce score a été comparé à une norme de référence de TB pulmonaire active déclarée par analyse de courbe de la fonction d'efficacité du récepteur (ROC) et de localisation ROC (LROC).

RÉSULTATS:: Aprés analyse ROC, è une sensibilité de 95%, la spécificité a été de 55,71% (IC95% 55,21–56,20) et la valeur prédictive négative a été de 99,98% (IC95% 99,95–99,99). La zone sous la courbe ROC a été de 0,90 (IC95% 0,86–0,93). Les résultats après analyse LROC ont été similaires.

CONCLUSION:: Le logiciel peut identifier plus de la moitié des images normales dans un contexte de dépistage de la TB, tout en préservant la sensibilité élevée, et pourrait donc être utilisé pour le triage.

MARCO DE REFERENCIA:: Los programas de cribado de la tuberculosis (TB) se optimizan cuando se reduce el número de las radiografías de tórax (CXR) que exigen una interpretación por personas expertas.

OBJETIVO:: Evaluar la eficacia de un programa informético de detección que clasifique las CXR, en el marco de un programa de cribado de la TB informético ultrarrépido y portétil.

MÉTODO:: Se evaluó de manera retrospectiva el programa informético a partir de una base de datos de 38 961 CXR posteroanteriores realizadas del 2005 al 2010 en pacientes únicos, de los cuales en 87 se diagnosticó TB. El programa generaba una puntuacióon de probabilidad de TB en cada CXR. Esta puntuación se comparó con una norma de referencia de la TB pulmonar activa notificada, mediante un anélisis de rendimiento diagnóstico de la característica operativa del receptor (ROC) y un anélisis ROC de localización (LROC).

RESULTADOS:: Tras el anélisis ROC, con una sensibilidad de 95%, la especificidad fue 55,71% (IC95% 55,21–56,20) y el valor diagnóstico de un resultado negativo fue 99,98% (IC95% 99,95–99,99). El área bajo la curva de rendimiento diagnóstico fue 0,90 (IC95% 0,86–0,93). Los resultados del anélisis LROC fueron equivalentes.

CONCLUSIÓN:: El programa informético puede detectar més de la mitad de las CXR normales en el contexto del cribado de la TB y al mismo tiempo conserva una alta sensibilidad, por lo cual se puede utilizar en la selección radiogréfica.

PubMed Disclaimer

Conflict of interest statement

Conflict of interest: JM and RHHMP are employees of Thirona, Nijmegen, The Netherlands, which develops CAD4TB software. BvG is co-founder and shareholder of Thirona. The remaining authors report no conflicts.

Figures

Figure 1.
Figure 1.
Proposed workflow with integration of CAD for tuberculosis detection as a first triage test before human reading.* Predetermined cut-off value.† Clinical and microbiological assessment. CXR = chest radiograph; CAD = computer-aided detection.
Figure 2.
Figure 2.
ROC and LROC curve analyses of CAD performance in discriminating between active TB and non-active TB cases. The triage cut-off point at 95% sensitivity is indicated by an arrow. Cases to the left of the cut-off point would be referred for human reading, whereas cases to the right would be excluded from further analysis. ROC = receiver operating characteristic; LROC = localisation ROC; CAD = computer-aided detection; TB = tuberculosis.
Figure 3.
Figure 3.
Examples of CAD analysis. First column: input CXR; second column: heat map resulting from texture analysis (colours indicate low-to-high suspicion of abnormality in the following order: blue-green-yellow-orange-red), last column: TB score assigned by CAD ranging from 0 (normal) to 100 (abnormal). CXR = chest radiograph; TB = tuberculosis; CAD = computer-aided detection. This image can be viewed online in colour at http://www.ingentaconnect.com/content/iuatld/ijtld/2018/00000022/00000005/art000...
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

See all "Cited by" articles

References

    1. World Health Organization. . Global tuberculosis report, 2016. WHO/HTM/TB/2016.13 Geneva, Switzerland: WHO, 2016.
    1. de Vries G, van Hest R A H, Richardus J H.. Impact of mobile radiographic screening on tuberculosis among drug users and homeless persons. Am J Respir Crit Care Med 2007; 176: 201– 207. - PubMed
    1. Story A, Murad S, Roberts W, Verheyen M, Hayward A C.. Tuberculosis in London: the importance of homelessness, problem drug use and prison. Thorax 2007; 62: 667– 671. - PMC - PubMed
    1. Steingart K R, Schiller I, Horne D J, Pai M, Boehme C C, Dendukuri N.. Xpert MTB/RIF assay for pulmonary tuberculosis and rifampicin resistance in adults. Cochrane Database Syst Rev 2014; 1: CD009593. - PMC - PubMed
    1. Shamshirband S, Hessam S, Javidnia H, . et al. Tuberculosis disease diagnosis using artificial immune recognition system. Int J Med Sci 2014; 11: 508– 514. - PMC - PubMed

Publication types

LinkOut - more resources