Table 2.
Characteristics of the radiomics research for microvascular invasion (MVI) prediction.
| Study ID | Imaging Modality | Phase/Sequence | Segmentation | Extension of VOI | Feature Selection | Number of Imaging Features Included in the Model | Event/Feature Ratio during Model Development | Clinical Variables for Modeling | AUC in Test Cohort |
|---|---|---|---|---|---|---|---|---|---|
| Zheng 2017 | CE-CT | PVP | 2D; auto | Yes (5-pixel) |
univariable logistic regression | 21 | 1.6 | AFP, tumor size, hepatitis | NA |
| Peng 2018 | CE-CT | AP, PVP, DP | 3D; semi-auto | No | LASSO | 8 | 15.9 | AFP, non-smooth tumor margin, internal arteries, hypoattenuating halos | 0.84 |
| Ma 2018 | CE-CT | AP, PVP, DP | 3D; manually | No | ICC/CCC, LASSO | 7 | 5.3 | Age, tumor size, hepatitis B | 0.80 |
| Feng 2019 | Gd-EOB-DTPA MRI | HBP | 3D; manually | Yes (10 mm) |
LASSO | 10 | 4.2 | NA | 0.84 |
| Ni 2019 | CE-CT | PVP | 2D; manually | No | LASSO, neighbourhood rough set, PCA | Unclear | NA | NA | NA |
| R. Zhang 2019 | Multimodel MRI | AP, PVP, EP, T1, T2, DWI | 3D; manually | Yes (10 mm) |
mRMR | 12 | 5.3 | AFP, arterial peritumoral enhancement | 0.86 |
| Zhu 2019 | CE-MRI | AP, PVP | 3D; manually | No | Kruskal-Wallis test; Pearson correlation | 4 | 9.3 | AFP, tumor size, differentiation | 0.79 |
| Nebbia 2020 | CE-MRI | AP, PVP, T1, T2, DWI | 3D; manually | Yes (10 pixel) |
LASSO | 17 | 3.6 | NA | NA |
| Q. Liu 2020 | CE-CT | AP, PVP | 3D;semiauto | No | ICC, RF | 28 | 5.3 | NA | 0.79 |
| X. Zhang 2020 | CE-CT | DP | 3D; manually | Yes (10 mm) |
LASSO | 44 | 4.0 | Age, AFP | 0.80 (0.80 in the validation cohort) |
| Jiang 2020 | CE-CT | AP, PVP, DP | 3D; manually | Yes (10 mm) |
Xgboost/3D-CNN | Unclear | NA | AFP | 0.91 |
| He 2020 | CE-CT | PVP | 3D; auto | No | LASSO | 2 | 35.5 | AFP, neutrophilic granulocytes, hemoglobin | 0.71 (0.84 in the validation cohort) |
| Chong 2021 | Gd-EOB-DTPA MRI | AP, PVP, HBP, DWI | 3D; manually | Yes (5 mm,10 mm, 50%) |
ICC, LASSO, Univariate Feature Selection | 74 | 0.9 | AFP, TBIL, capsule enhancement, peritumoral enhancement | 0.92 |
| Chen 2021 | Gd-EOB-DTPA MRI | AP, PVP, HBP, T1, T2, DWI | 3D; manually | No | LASSO | 21 | NA | NA | 0.94 |
| Li 2021 | PET/CT | [18F]FDG PET/CT | 3D; semiauto | No | LASSO | 11 | 0.7 | NA | 0.69 |
| Song 2021 | CE-MRI | AP, PVP, DP, T1,T2, ADC, DWI | 3D; manually | No | PCA, analysis of variance | Unclear | NA | NA | 0.73 |
| Dai 2021 | Gd-EOB-DTPA MRI | AP, PVP, HBP, T1 | 3D; manually | No | LASSO-RFE, LASSO, mRMR, SVM-RFE | 5 | 5.8 | NA | 0.90 (LOOCV) |
| P. Liu 2021 | CE-CT | AP | 3D; manually | No | ICC, LASSO | 10 | 4.2 | NA | 0.75 |
| Sh. Zhang 2021 | Gd-EOB-DTPA MRI | HBP (5, 10, 15 min) | 3D; manually | No | LASSO | 14 | NA | NA | NA |
| W. Zhang 2021 | CE-CT | EAP, LAP, PVP, EP | 3D; manually | 2, 4, 6, 8, 10, 12, 14 mm | 15 methods (Fisher score, t score, etc.) | Unclear | NA | NA | 0.81 |
| Meng 2021 | Multiparametric MRI & multiphasic CT | AP, PVP, T2, DWI (MRI)/AP, PVP (CT) | 3D; manually | Yes (3 mm) |
ICC, univariate analysis, feature reduction, LASSO | 16/16 | 8.1/1.9 # | NA | 0.80 |
| Y. Zhang 2021 | Multiparametric MRI | AP, PVP, DP, T2, DWI, ADC | 3D; manually | No | ICC, analysis of variance, Mann-Whitney U-test, correlation analysis, LASSO | 47 | 1.6 | Age, AFP, tumor size | 0.84 |
Note: # for CT and MRI model respectively. 3D-CNN, 3D-convolutional neural network; ADC, apparent diffusion coefficient; AFP, alpha-fetoprotein; AP, arterial phase; CCC, concordance correlation coefficient; CE-CT, contrast enhanced-CT; CE-MRI, contrast enhanced-MRI; DP, delay phase; DWI, diffusion-weighted imaging; EAP, early arterial phase; EP, equilibrium phase; HBP, hepatobiliary phase; ICC, interclass correlation coefficient; LASSO, least absolute shrinkage and selection operator; LOOCV, leave-one out cross validation; mRMR, minimum redundancy-maximum relevance; NA, not applicable; PCA, principal component analysis; PVP, portal vein phase; RFE, recursive feature elimination; SVM, Support vector machine; TBIL, total bilirubin; VOI, volume of interest.