The value of CT-based radiomic signature in predicting lumbar osteoporosis in the elderly
JIANG Lezhen1,GUO Yifan2, SU Jiehui1, LIN Wenxiao1, WU Aiqin1, BAI Guanghui1
1.Department of Radiography, the Second Affiliated Hospital of Wenzhou Medical University, Wenzhou 325027, China; 2.Department of Radiography, the First Affiliated Hospital of Zhejiang Chinese Medical University, Hangzhou 310006, China
JIANG Lezhen,GUO Yifan,SU Jiehui, et al. The value of CT-based radiomic signature in predicting lumbar osteoporosis in the elderly[J]. JOURNAL OF WEZHOU MEDICAL UNIVERSITY, 2021, 51(11): 885-890.
Abstract：Objective: To evaluate the value of radiomic signature based on CT plain scan in predicting lumbar osteoporosis in the elderly. Methods: Sixty-six cases including postmenopausal women and men over 50 years old in the Second Affiliated Hospital of Wenzhou Medical University from July 2020 to October 2020 were studied. All patients received CT plain scan of lumbar spine, of whom 237 lumbar vertebrae from lumbar 1 to lumbar 4 met the study criteria, with 95 osteoporosis and 142 non-osteoporosis. All lumbar vertebrae were stratified randomly into the training group and the validation group in a ratio of 7:3. The cancellous bone of the lumbar vertebrae was delineated layer by layer on CT images using 3D Slicer software to obtain the three-dimensional region of interest (ROI), and the radiomic features were extracted from each lumbar vertebra. Radiomic features of the training group were dimensionality reduced by the Min-redundancy and Max Relevance (mRMR) and 10 features were retained. The least absolute shrinkage and selection operator (LASSO) logistic regression was then applied to select the optimized subset from the retained features to construct the radiomic signature. The area under the receiver (AUC) of operating characteristic (ROC) curve was used to evaluate the predictive efficacy of the radiomic signature. Calibration curves were used to determine the calibration of the radiomic signature.Results: Of the 1 316 radiomic features extracted, 7 radiomic features significantly associated with lumbar osteoporosis were selected and used to construct radiomic signature. The ROC curves showed that the AUC of the radiomic signature in the training and validation groups were 0.908 (95%CI=0.863-0.952) and 0.935(95%CI=0.873-0.996), respectively. Calibration curves showed that the radiomic signature had good fitting degree in both the training and validation group. Conclusion: The CT-based radiomic signature can be used as a noninvasive tool to evaluate elderly patients with lumbar osteoporosis, which may be helpful to assist clinical decision-making and improve patient prognosis.
