| Objective:To investigate the differential diagnostic value of multiple myeloma and osteolytic metastases based on lumbar vertebral MRI radiomics nomogram.Methods:273 patients with lumbar multiple myeloma（MM）and osteolytic metastasis（OM）admitted to the Affiliated Hospital of Qingdao University from June 2008 to June2022（training set=190,validation set=83）were enrolled into this study.Pretreatment conventional MRI images and clinical information of all patients were complete.Independent risk factors for differential diagnosis of lumbar MM and OM were screened based on clinical information and convention MRI features,and clinical models were constructed to verify their differential diagnosis efficacy.The minimum redundancy maximum relevance algorithm and least absolute shrinkage and selection operator algorithm were adopted to screen radiomics features,followed by 4 radiomics signatures（RS）constructing:（1）an RS based on T1WI（T1WI-RS）,（2）an RS based on T2WI（T2WI-RS）（3）an RS based on FS-T2WI（FS-T2WI-RS）（4）an RS based on the combination of T1WI,T2WI and FS-T2WI（Combined-RS）.Based on the RS model with best performance（Combined-RS）and clinical model,we constructed a radiomics nomogram model for identification of lumbar MM and OM.Area under the receiver operating characteristics curve（AUC）,calibration curves,and decision curve analysis were applied to assess the performance and clinical applicability of the models.Result:Univariate-multivariate logistic regression analysis of clinical information and conventional MRI features showed that pedicle involvement,T1WI signal heterogeneity,FS-T2WI signal heterogeneity and vertebral segment involvement were identified as risk factors with differential diagnostic value.The performance of clinical model based on these risk factors was unconvincing for differential diagnosis of lumbar MM and OM,with an AUC of 0.828[95%confidence interval（CI）:0.0.771-0.884]and accuracy of 72.1%in the training set and an AUC of 0.686（95%CI:0.573-0.799）and accuracy of 62.7%in the validation set.Among four RS-models,the Combined-RS model achieved the best differential diagnostic efficacy,with an AUC of 0.969（95%CI:0.946-0.992）and accuracy of 92.1%in the training set and an AUC of 0.856（95%CI:0.777-0.935）and accuracy of 77.1%in the validation set,which is higher than that of T1WI-RS model（AUC of 0.939,95%CI of 0.906-0.971 and accuracy of 85.8%in the training set,and AUC of 0.729,95%CI of 0.620-0.838 and accuracy of 63.9%in the validation set）,T2WI-RS model（AUC of 0.934,95%CI of 0.899-0.968 and accuracy of 87.9%in the training set,AUC of 0.781,95%CI of 0.684-0.879 and accuracy of 67.5%in the validation set）and FS-T2WI-RS model（AUC of 0.914,95%CI of 0.874-0.954 and accuracy of 85.3%in the training set,AUC of 0.830,95%CI of 0.774-0.916 and accuracy of 72.3%in the validation set）.The radiomics nomogram model achieved the best differential diagnostic efficacy with an AUC of 0.980（95%CI:0.0.777-0.996）and an accuracy of 94.7%in the training set and an AUC of 0.860（95%CI:0.777-0.944）and an accuracy of 77.3%in the validation set,which is higher than that of the clinical model and the best-performing RS model（Combined-RS model）.The calibration curves displayed the radiomics nomogram achieved excellent calibration effect in both the training set and the validation set.The decision curve analysis showed the radiomics nomogram model have greater clinical applicability than the other two models.Conclusion:Lumbar vertebral MRI based radiomics nomogram has a convincing differential diagnostic performance in MM and OM,and can provide effective information for the formulation of clinical treatment strategy. |
