.png){kind=icon}
.png){kind=icon}
Return
Active tuberculosis screening incorporated with AI-assisted chest X-ray among the elderly in community settings, a cross-sectional study in China
.png){kind=icon}
OA
AI
C
Q
W
H
Q
C
L
DOI：10.1186/s12877-026-08113-2.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Older adults face an elevated risk of tuberculosis (TB), yet they frequently present with atypical or asymptomatic clinical features, leading to major diagnostic delays. This study evaluated the feasibility, diagnostic yield, and epidemiological insights of a novel “double X plus AI” active case-finding (ACF) strategy embedded within the routine annual health check-up program for community-dwelling elderly. This cross-sectional study was conducted across 14 communities in Jiangsu Province. Local residents aged ≥ 65 years received symptom screening and digital chest X-ray for tuberculosis by annual health examination. Chest X-rays were triaged using a convolutional neural network-based artificial intelligence (AI) system. Participants with tuberculosis-related symptoms or AI-suspected chest X-ray findings provided sputum specimens for molecular analysis using an 8:1 pooling strategy via the Xpert® MTB/RIF Ultra assay. Positive pools were subsequently disaggregated for individual testing. A total of 14,312 older adults were enrolled, among whom 2,436 presumptive tuberculosis patients were identified, comprising 2,309 AI-suspected individuals and 127 symptomatic individuals with normal chest X-rays. Among the 2,309 AI-suspected individuals, 76 failed to provide qualified sputum specimens. A total of 20 active pulmonary TB cases were disclosed, including 16 microbiological positive and 4 clinical diagnosed tuberculosis patients. The overall age and sex standardized prevalence of pulmonary tuberculosis identified through this active screening strategy was 154.5 per 100,000, which was higher than the registered prevalence among older adults in Jiangsu Province in 2025 derived from passive surveillance (48.9/100,000; P < 0.001). Notably, all the twenty confirmed patients were originated from the AI-suspected group, while none of the symptomatic individuals with normal X-rays were diagnosed with tuberculosis. Among the confirmed cases, 70.0% (14/20) were asymptomatic, and 80.0% (16/20) were microbiologically positive. The median AI-derived TB suspicion score for these confirmed cases was 0.89 (IQR: 0.83–0.95), with all cases scoring above 0.73. Crucially, these AI scores demonstrated no significant difference from both symptomatic status (P = 0.710) and microbiological outcomes (P = 0.813). Embedding an AI-triaged, pooled molecular testing workflow into existing geriatric public health infrastructure is feasible and effective for identifying tuberculosis among community-dwelling older adults. This symptom-agnostic strategy identified a substantial burden of asymptomatic tuberculosis that may be missed by conventional symptom-based screening, highlighting the value of integrating AI-assisted active screening into routine health services for high-risk older populations.
Keywords:
Tuberculosis
Active case finding
X-ray
Pooled sputum testing
Artificial intelligence
Journal
IF:
3.8
Papers:
7.9K
Citations:
2.4W
Organization
Cited Papers