Multisequence conventional MRI-based habitat radiomics for differentiating Brucella and tuberculosis spondylitis: a dual-center study

Brucella spondylitis (BS) and tuberculosis spondylitis (TS) are clinically important causes of infectious spondylitis with substantial overlap in symptoms, laboratory findings, and conventional MRI appearances, often delaying etiologic diagnosis and appropriate therapy. Although MRI is sensitive for detecting spinal infection, qualitative interpretation lacks specificity. Radiomics offers quantitative assessment, but whole-lesion approaches may overlook intralesional heterogeneity. Multisequence conventional MRI-based habitat radiomics may better capture spatially heterogeneous MRI signal phenotypes to improve BS–TS differentiation. This retrospective dual-center study enrolled 353 patients with spinal infections. Training and validation cohorts comprised 303 cases from Center 1, with an independent test set of 50 cases from Center 2. MRI sequences (T1WI, T2WI, fat-suppressed T2WI) underwent standardized preprocessing and nnU-Net segmentation; external-test masks were reviewed and corrected by a radiologist blinded to diagnosis when necessary. Radiomic features were extracted and feature selection was performed via ANOVA and LASSO regression. Habitat analysis using k-means clustering identified informative features characterizing intralesional heterogeneity. Radiomics, habitat, and combined models were developed using machine learning classifiers and evaluated using AUC, sensitivity, specificity, and decision curve analysis. A total of 6,792 radiomic features were extracted. Seventeen radiomic features and 10 habitat features were selected. The radiomics model achieved AUCs of 0.872 (95% CI 0.827–0.917), 0.875 (0.785–0.965), and 0.825 (0.711–0.939) in the training, internal model-selection, and external test cohorts, respectively. Corresponding AUCs were 0.912 (0.878–0.946), 0.860 (0.766–0.954), and 0.801 (0.637–0.966) for the habitat model and 0.918 (0.883–0.953), 0.885 (0.797–0.973), and 0.842 (0.699–0.985) for the combined model. In the external test cohort, the combined model had sensitivity of 0.971 (95% CI 0.903–1.000) and specificity of 0.625 (0.400–0.857). The combined model did not significantly improve AUC over either component model in the internal model-selection or external test cohort. Exploratory NRI and IDI analyses favored the combined model; these findings should be interpreted cautiously given the small external cohort and the absence of a clinical comparator. Combining whole-lesion radiomics with habitat-derived MRI signal features showed promising discrimination between chronic BS and TS classified using surgical histopathology in a surgically treated cohort. These findings are preliminary and require prospective multicenter validation in broader clinical populations before clinical use.

Authors

Institutions

Publication Details

Journal
European journal of medical research
Published
2026-11-17
DOI
https://doi.org/10.1186/s40001-026-05486-3
Primary Topic
Radiomics and Machine Learning in Medical Imaging
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Multisequence conventional MRI-based habitat radiomics for differentiating Brucella and tuberculosis spondylitis: a dual-center study

Abudouresuli Tuersun, Juan Tan, Maimaitiming Mahemuti, Haihong Ma et al.
European journal of medical research
Radiomics and Machine Learning in Medical Imaging
article

Multisequence conventional MRI-based habitat radiomics for differentiating Brucella and tuberculosis spondylitis: a dual-center study

Abudouresuli Tuersun, Juan Tan, Maimaitiming Mahemuti, Haihong Ma, Muyassar Gheni, Yang Li
article en

Abstract

Brucella spondylitis (BS) and tuberculosis spondylitis (TS) are clinically important causes of infectious spondylitis with substantial overlap in symptoms, laboratory findings, and conventional MRI appearances, often delaying etiologic diagnosis and appropriate therapy. Although MRI is sensitive for detecting spinal infection, qualitative interpretation lacks specificity. Radiomics offers quantitative assessment, but whole-lesion approaches may overlook intralesional heterogeneity. Multisequence conventional MRI-based habitat radiomics may better capture spatially heterogeneous MRI signal phenotypes to improve BS–TS differentiation. This retrospective dual-center study enrolled 353 patients with spinal infections. Training and validation cohorts comprised 303 cases from Center 1, with an independent test set of 50 cases from Center 2. MRI sequences (T1WI, T2WI, fat-suppressed T2WI) underwent standardized preprocessing and nnU-Net segmentation; external-test masks were reviewed and corrected by a radiologist blinded to diagnosis when necessary. Radiomic features were extracted and feature selection was performed via ANOVA and LASSO regression. Habitat analysis using k-means clustering identified informative features characterizing intralesional heterogeneity. Radiomics, habitat, and combined models were developed using machine learning classifiers and evaluated using AUC, sensitivity, specificity, and decision curve analysis. A total of 6,792 radiomic features were extracted. Seventeen radiomic features and 10 habitat features were selected. The radiomics model achieved AUCs of 0.872 (95% CI 0.827–0.917), 0.875 (0.785–0.965), and 0.825 (0.711–0.939) in the training, internal model-selection, and external test cohorts, respectively. Corresponding AUCs were 0.912 (0.878–0.946), 0.860 (0.766–0.954), and 0.801 (0.637–0.966) for the habitat model and 0.918 (0.883–0.953), 0.885 (0.797–0.973), and 0.842 (0.699–0.985) for the combined model. In the external test cohort, the combined model had sensitivity of 0.971 (95% CI 0.903–1.000) and specificity of 0.625 (0.400–0.857). The combined model did not significantly improve AUC over either component model in the internal model-selection or external test cohort. Exploratory NRI and IDI analyses favored the combined model; these findings should be interpreted cautiously given the small external cohort and the absence of a clinical comparator. Combining whole-lesion radiomics with habitat-derived MRI signal features showed promising discrimination between chronic BS and TS classified using surgical histopathology in a surgically treated cohort. These findings are preliminary and require prospective multicenter validation in broader clinical populations before clinical use.

European journal of medical research
United Imaging Healthcare (China) (CN), Kashi University (CN)
Good health and well-being
Radiomics and Machine Learning in Medical Imaging
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon