Application Of Radiomics Based On MRI In Differential Diagnosis Of Tuberculous Spondylitis And Brucella Spondylitis

Objective: To investigate the value of machine learning(ML)model based on conventional MRI radiomics in the differential diagnosis of tuberculous spondylitis(TS)and Brucella spondylitis(BS).Methods: Routine MRI images of patients with 118 cases of TS and 85 cases of BS confirmed by clinical diagnosis or postoperative pathology were retrospectively collected,including sagittal T1 WI,T2WI and FS-T2 WI images on plain scan.The patient images(in DICOM format)and clinical data were imported into Huiyi Huiying big data scientific research platform,and ROI was drawn layer by layer.They were randomly assigned to the training group(n=162)and the validation group(n=41)at a ratio of 8:2.Radiomics features of lesions in T1 WI,T2WI and FS-T2 WI sequences were extracted.Then,three methods including Select K Best,Variance Threshold and LASSO were used to screen and reduce the dimension of radiomics features.Further,six machine learning methods including LR,SVM,DT,RF,XG Boost and KNN were used to analyze and construct omics models based on T1 WI,T2WI and FS-T2 WI sequences,and ROC curves were drawn to evaluate the diagnostic performance of each model.Results: A total of 1409 radiomics features were extracted based on T1 WI,T2WI and FS-T2 WI models,and 25,14 and 9 best radiomics features were obtained by three dimensionality reduction methods,respectively.In the training set,the AUC,95% confidence interval,sensitivity and specificity of LR classifiers based on T1 WI and T2 WI,and SVM classifiers based on FS-T2 WI model in each model were0.913,0.857-0.969,0.840 and 0.850,respectively.0.821,0.753-0.889,0.730,0.750;0.897,0.834-0.960,0.790,0.790.In the test set,the AUC,95%CI,sensitivity and specificity of LR classifiers based on T1 WI and T2 WI,and SVM classifiers based on FS-T2 WI model in each model were 0.760,0.603-0.917,0.670 and 0.590,respectively.0.924,0.799-1.000,0.830,0.760;0.801,0.659-0.943,0.750,0.650.The F1 score,precision and recall of LR classifier based on T2 WI model in the training set were 0.830,0.830 and 0.830,respectively.Conclusion: LR classifiers based on T1 WI and T2 WI,and SVM classifiers based on FS-T2 WI model have the best differential diagnostic performance among all models.Radiomics based on T1 WI,T2WI and FS-T2 WI images can distinguish TS from BS.The LR classifier based on T2 WI model is the most stable and has the best discriminative performance for TS and BS.