The feasibility of radiomics model in opportunistic screening of three-classification bone condition on chest CT images
10.3969/j.issn.1002-1671.2025.07.031
- VernacularTitle:胸部CT图像构建三分类骨质状态筛查影像组学模型可行性研究
- Author:
Changyu DU
1
;
Yijun LIU
1
;
Shigeng WANG
1
;
Xiaoyu TONG
1
;
Wei WEI
1
;
Anliang CHEN
1
;
Qiye CHENG
1
Author Information
1. 大连医科大学附属第一医院放射科,辽宁 大连 116011
- Publication Type:Journal Article
- Keywords:
osteoporosis;
radiomics;
opportunistic screening;
bone mineral density;
computed tomography
- From:
Journal of Practical Radiology
2025;41(7):1220-1224
- CountryChina
- Language:Chinese
-
Abstract:
Objective To explore the feasibility of constructing a three-classification bone status screening radiomics model on chest CT images.Methods A total of 371 patients who underwent both chest and abdominal plain CT examinations were retrospec-tively selected and randomly divided into training set(296 cases)and test set(75 cases)in a ratio of 8︰2.Additionally,110 patients were included as external validation set using the same criteria.The 120 kVp abdominal images were transmitted to a quantitative compu-ted tomography(QCT)post-processing workstation to measure the bone mineral density(BMD)of the L1-L2 vertebral bodies.Patients were classified into osteoporosis(OP)group(BMD<80 mg/cm3),osteopenia group(80 mg/cm3≤BMD≤120 mg/cm3)and normal bone mass group(BMD>120 mg/cm3)based on QCT BMD results.The automatic segmentation model was used to segment T10-T12 vertebral trabecular bone on chest CT images and the radiomics models based on random forest(RF)and logistic regres-sion(LR)was established to evaluate BMD,enabling it to simultaneously distinguish OP,osteopenia,and normal bone mass.The diag-nostic performance of the two models were evaluated using metrics such as the area under the curve(AUC),sensitivity and specificity.The DeLong test was used to compare the differences between the two models.Results In the test set,the AUC for differentiating normal bone mass were 0.948 and 0.877 for the RF and LR models,respectively;the AUC for differentiating OP were 0.942 and 0.836,respectively;and the AUC for differentiating osteopenia were 0.871 and 0.688,respectively.The performance comparison results of the models showed that there was no statistically significant difference in AUC(0.966 vs 0.907,P>0.05)between RF model and LR model in the external validation set for distinguishing OP,while there was a statistically significant difference in AUC for distinguishing osteopenia(0.895 vs 0.749,P=0.009)and normal bone mass(0.975 vs 0.906,P=0.023).The RF model performance was superior to the LR model.Conclusion The radiomics model developed based on chest plain CT can be used for opportunistic OP screening with good diagnostic efficacy,and the the model based on the RF classifier outperforms the LR model.
