Preliminary Study On The Graded Diagnosis Of Osteoporosis Based On Quantitative CT Of Radiomics

Objective: To investigate the classification and diagnosis of osteoporosis based on quantitative CT radiomics features,and the diagnostic efficacy of different diagnostic models was evaluated by AUC,accuracy,sensitivity,and specificity to distinguish osteoporosis from bone loss,so as to better guide clinical judgment and treatment.Materials and methods: This is a retrospective study of a total of 635 patients who underwent quantitative CT from November 2016 to November 2019 at Dazhou Central Hospital.The samples were divided into normal group(T value ≥-1),osteopenia group(-1 < T value <-2.5)and osteoporosis group(-2.5 ≤ T value).After the screening and exclusion,vertebral cancellous bone segmentation was performed on the CT scan images using 3D-Slicer software,and the acquired imaging histological parameters were selected for feature descending and peacekeeping using the minimum absolute shrinkage and selection operator(LASSO)and the minimum-redundancy maximum correlation(m RMR).The radiomics features with higher classification value were evaluated.Finally,the multivariate logistic regression was applied to construct the classification models including radiomics model,clinics model,and the combined radiomics model(CRM).The area under the curve(AUC),accuracy,specificity,sensitivity,positive predictive value and negative predictive value were used to evaluate the performance of the models.Results: Among the 851 image features extracted from quantitative CT scans,6optimal classification features were screened.A radiomics model was developed based on these features.Using tenfold cross-validation,patients in the osteoporosis and osteopenia groups were divided into a training set(N=414)and a validation set(N=176).13 features including age,uric acid,alkaline phosphatase(ALP),and homocysteine(HCY)were screened as predictors.Three clinical indicators,including age,ALP,and HCY,were used to construct the clinical feature model and CRM for osteoporosis and osteopenia.The AUC of CRM was 0.96(95% CI,0.95-0.98)in the training cohort and0.96(95% CI,0.92-1.00)in the validation cohort,which was better than the clinical characteristics model(AUC=0.81 in the training cohort and AUC=0.79 in the validation cohort),but not superior to the radiomics model alone(AUC of 0.96 in the validation cohort,95% CI,0.92-1.00).Calibration curves indicated that the predicted values of the radiomics and nomogram were in good accordance with the observed values,and decision curve analysis demonstrated the clinical application of the radiomics model.Conclusions: The radiomics features based on QCT of the lumbar spine have excellent efficacy in grading the diagnosis of osteoporosis and are in close conformity with the clinical diagnosis.The conjunction of 3 features of age,alkaline phosphatase and homocysteine had the ability to differentiate osteoporosis from osteopenia.And the stability of the combined model was lessened by the enrollment of the clinical feature cohort.