A systematic review and meta-analysis of the diagnostic accuracy of artificial intelligence in detecting tuberculosis using cough sounds
1
Health Technology Assessment, ICMR - Regional Medical Research Centre,Bhubaneswar, Odisha, Bhubaneswar, India
Popul. Med. 2026;8(Supplement Supplement 1):A877
ABSTRACT
BACKGROUND:
Tuberculosis (TB) remains the leading cause of death from infectious diseases globally, with significant diagnostic challenges in low- and middle-income countries. Artificial intelligence analysis of cough sounds could offer an inexpensive and accessible solution for detecting TB. This systematic review and meta-analysis evaluated the diagnostic accuracy of AI-based cough sound analysis for screening TB and identified key methodological gaps.
METHODS:
We performed a systematic review (PROSPERO: CRD420250656065), searching PubMed, Scopus, IEEE, Web of Science, and CINAHL for studies published between 1 January 2009 and 31 December 2024. Eligible studies involved the development or application of artificial intelligence algorithms for tuberculosis detection based on cough sound analysis. Risk of bias was assessed using the QUADAS-AI tool. Binary diagnostic accuracy data, including sensitivity, specificity, and area under the curve (AUC), were extracted to quantify diagnostic performance.
RESULTS:
Overall, fourteen studies were found, largely from Asia & Africa. Meta-analysis of seven studies showed a pooled sensitivity of 91% (95% CI: 88–94%) and specificity of 89% (95% CI: 85–92%), with a diagnostic odds ratio of 81.61 and an area under the curve of 0.9539, indicating strong diagnostic accuracy. Superior accuracy was displayed by deep learning models, having a 92% (95%: 88–96%) sensitivity and 91% (95%: 86–94%) specificity, while machine learning models achieved slightly lower but still reasonable 89% (95%: 87–91%) sensitivity and 84% (95%: 76–90%) specificity. Sample size subgroup analysis by ≤1000 or >1000, and geographic region by Asia or non-Asia, revealed uniformly satisfactory diagnostic performance across strata.
CONCLUSIONS:
Artificial intelligence models hold a lot of promise as a non-invasive, rapid screening tool for tuberculosis detection, particularly in resource-limited settings. However, the high risk of bias, heterogeneity, and reliance on internal validation highlights the need for large-scale, externally validated, multi-centre studies before clinical adoption.
Tuberculosis (TB) remains the leading cause of death from infectious diseases globally, with significant diagnostic challenges in low- and middle-income countries. Artificial intelligence analysis of cough sounds could offer an inexpensive and accessible solution for detecting TB. This systematic review and meta-analysis evaluated the diagnostic accuracy of AI-based cough sound analysis for screening TB and identified key methodological gaps.
METHODS:
We performed a systematic review (PROSPERO: CRD420250656065), searching PubMed, Scopus, IEEE, Web of Science, and CINAHL for studies published between 1 January 2009 and 31 December 2024. Eligible studies involved the development or application of artificial intelligence algorithms for tuberculosis detection based on cough sound analysis. Risk of bias was assessed using the QUADAS-AI tool. Binary diagnostic accuracy data, including sensitivity, specificity, and area under the curve (AUC), were extracted to quantify diagnostic performance.
RESULTS:
Overall, fourteen studies were found, largely from Asia & Africa. Meta-analysis of seven studies showed a pooled sensitivity of 91% (95% CI: 88–94%) and specificity of 89% (95% CI: 85–92%), with a diagnostic odds ratio of 81.61 and an area under the curve of 0.9539, indicating strong diagnostic accuracy. Superior accuracy was displayed by deep learning models, having a 92% (95%: 88–96%) sensitivity and 91% (95%: 86–94%) specificity, while machine learning models achieved slightly lower but still reasonable 89% (95%: 87–91%) sensitivity and 84% (95%: 76–90%) specificity. Sample size subgroup analysis by ≤1000 or >1000, and geographic region by Asia or non-Asia, revealed uniformly satisfactory diagnostic performance across strata.
CONCLUSIONS:
Artificial intelligence models hold a lot of promise as a non-invasive, rapid screening tool for tuberculosis detection, particularly in resource-limited settings. However, the high risk of bias, heterogeneity, and reliance on internal validation highlights the need for large-scale, externally validated, multi-centre studies before clinical adoption.
| eISSN: | 2654-1459 |
We process personal data collected when visiting the website. The function of obtaining information about users and their behavior is carried out by voluntarily entered information in forms and saving cookies in end devices. Data, including cookies, are used to provide services, improve the user experience and to analyze the traffic in accordance with the Privacy policy. Data are also collected and processed by Google Analytics tool (more).
You can change cookies settings in your browser. Restricted use of cookies in the browser configuration may affect some functionalities of the website.
You can change cookies settings in your browser. Restricted use of cookies in the browser configuration may affect some functionalities of the website.
