Related Experiment Video
Updated: Oct 29, 2026
Diagnosing Pulmonary Tuberculosis with the Xpert MTB/RIF Test
Published on: April 9, 2012
A Multicenter Ultrasound Habitat Radiomics Framework For Diagnosing Tuberculous Lymphadenitis And Stratifying
Tingting Qian1, Peijun Chen2, Yang Zhang3
1Department of Ultrasound, Affiliated Hangzhou First People's Hospital, School of Medicine, Westlake University, Hangzhou, China (T.Q., Y.Q., G.C., Y.W., G.Y.).
Rationale And Objectives:
To develop and validate an ultrasound-based habitat radiomics framework for the hierarchical differentiation of tuberculous lymphadenitis (TBLN) from non-tuberculous lymphadenopathy, and to noninvasively predict rifampicin resistance risk using an independently collected cohort of patients with confirmed TBLN.
Materials And Methods:
In this multicenter retrospective study, 618 patients with superficial lymphadenopathy who underwent ultrasound examination between January 2022 and December 2025 at four tertiary hospitals were included. Among them, 530 patients with definitive diagnoses were used for diagnostic differentiation and were divided into a model-development (training) cohort and an external validation cohort, while 236 patients with confirmed tuberculous lymphadenitis (TBLN) were included for rifampicin resistance risk assessment. Based on standardized B-mode ultrasound images, we developed a 2-stage ultrasound habitat radiomics framework that first differentiated tuberculous from non-tuberculous lymph nodes. Rifampicin resistance risk was assessed within the confirmed TBLN cohort using nested cross-validated out-of-fold predictions. Model performance was evaluated using receiver operating characteristic (ROC) analysis, calibration curves, and decision curve analysis (DCA), and model interpretability was assessed using SHAP-based methods.
Results:
For differentiating tuberculous from non-tuberculous lymph nodes, the habitat-based model demonstrated strong discriminative performance, achieving an area under the curve (AUC) of 0.83 in the training cohort and maintaining consistent performance in the external test cohort. Using nested cross-validated out-of-fold predictions within the confirmed TBLN cohort, the habitat-based model achieved an AUC of 0.83 for predicting rifampicin resistance, outperforming conventional radiomics and demonstrating superior clinical net benefit. Spatial visualization revealed that high-risk habitats were localized to specific intralesional regions, underscoring the importance of spatial heterogeneity in both disease diagnosis and drug resistance assessment.
Conclusion:
Ultrasound-based habitat radiomics provides an interpretable and clinically applicable framework for hierarchical diagnosis of tuberculous lymphadenitis and for rifampicin resistance risk assessment in an independent patient cohort, supporting precision imaging-guided management of tuberculosis.
