Table 2.
Radiomic characteristics of studies included in the systematic review.
| Author | Segmentation method | Radiomic feature categories | Machine-learning method for feature selection | Number of selected features | Model | AUC of radiomic model with the best performance | clinical factor | AUC of radiomic-clinical model | ||
|---|---|---|---|---|---|---|---|---|---|---|
| Training set | Validation set | Training set | Validation set | |||||||
| Xu | Semi-automatic segmentation(3D) | Signal intensity histogram-based features and 3D ND-Haralick texture features based intensity and its high-order derivative maps | SVM-RFE","SMOTE | 13 | SVM-RFE | 0.861 | NA | NA | NA | NA |
| Garapati | Automatic segmentation(3D) | First-order statistics, shape, contrast, GLRLM, | Stepwise feature selection | 3 subsets of radiomic features | LDA, NN, SVM, RF | 0.97 | NA | NA | NA | NA |
| Tong | Manual segmentation(3D) | LBP、GLCM | An optimal biomarker approach | 9 | SVM | Patient level:0.806,radial sector level:0.813 | NA | NA | NA | NA |
| Xu | Manual segmentation(2D) | Histogram, CM , RLM, | SVM-RFE","SMOTE, | 19 | SVM-RFE | 0.9857 | NA | NA | NA | NA |
| Zheng | Semi-automatic segmentation(3D) | first-order statistics,shape-based,GLCM,GLRLM,GLSZM,NGTDM,and GLDM | LASSO LR | 23 | LASSO | 0.913;Optimism-corrected:0.912 | 0.874 | Tumor size | 0.922;optimism-corrected AUC of 0.921 | 0.876 |
| Xu | Manual segmentation and automatic segmentation(3D) | First-order intensity features,high-order texture features,and shape ,GLCM,GLRLM,GLSZM and NGTDM | Boruta | 21 | RF, AR | .0.907 | 0.904 | RandomForest model and TURBT | NA | NA |
| Wang | Manual segmentation and automatic segmentation(2D) | Histogram , CM, RLM, NGTDM and GLSZM | SVM-RFE | 36 | LR, LASSO | 0.88 | external validation cohort 0.813 | Radscore and tumor stalk | 0.924 | 0.877 |
| Hammouda | Automatic segmentation(3D) | Histogram ,GLCM,GLRLM,and morphological features | NA | NA | NN(best)","RF,SVM | 0.9864 | NA | NA | NA | |
| Zhang | Semi-automatic segmentation(3D) | NA | NA | NA | FGP-Net | development cohort:0.936, tuning cohort:0.891 | internal validation cohort: 0.861,external validation cohort: 0.791 | NA | NA | NA |
| Zheng | manual segmentation (3D) | shape and size-based features, image intensity, textural features and wavelet features | mRMR | 40 | Lasso(best)、SVM、RF | 0.934 | 0.906 | VI-RADS | 0.97 | 0.943 |
| Zhou | semi-automatic segmentation(3D) | GLDM,Shape2D,GLCM,Shape3D,First-order,GLRLM,GLSZM,and NGTDM | SVM-RFE | 6 | LR, Decision Tree, SVM(best), and Adaboost algorithm | 0.898 | 0.702 | Rad-score, albuminuria and metabolic syndrome | 0.8457 | |
AR, all-relevant model; AUC, area under the curve; CM, Co-occurrence matrices; 3D, three dimensional; 2D, two dimensional; FGP-Net, Filter-guided Pyramid Network; GLCM, grey-level co-occurrence matrix; GLDM, gray level dependence matrix; GLRLM, gray-level run length matrix; GLSZM, gray-level size zone matrix; LASSO, least absolute shrinkage and selection operator; LBP, local binary pattern; LDA, linear discriminant analysis; LR, logistic regression; mRMR, min-redundancy; NA, not available; ND, nondirectional Haralic textural features; NN, neural network; NGTDM, neighborhood gray-tone difference matrix; RF, random forest model; RFE, recursive feature elimination; RLM, run length matrix; SMOTE, synthetic minority oversampling technique; SVM, support vector machine classififier; TURBT, transurethral resection of bladder tumor.