Table 3.
The performance of imaging, radiomics of single and multiple MR sequences, and final fusion models for predicting MVI in ICC patients.
| Models | Classifier and cohort | AUC | Accuracy | Sensitivity | Specificity | Precision |
|---|---|---|---|---|---|---|
| Imaging model | LR (TD/VD) | 0.726/0.522 | 0.669/0.545 | 0.605/0.400 | 0.696/0.609 | 0.451/0.308 |
| RF (TD/VD) | 0.742/0.578 | 0.731/0.697 | 0.211/0.100 | 0.946/0.957 | 0.615/0.500 | |
| SVM (TD/VD) | 0.726/0.483 | 0.708/0.697 | 0.000/0.000 | 1.000/1.000 | 0.000/0.000 | |
| DWI model | LR (TD/VD) | 1.000/0.530 | 1.000/0.485 | 1.000/0.600 | 1.000/0.435 | 1.000/0.316 |
| RF (TD/VD) | 0.943/0.530 | 0.800/0.697 | 0.316/0.000 | 1.000/1.000 | 1.000/0.000 | |
| SVM (TD/VD) | 1.000/0.774 | 1.000/0.697 | 1.000/0.000 | 1.000/1.000 | 1.000/0.000 | |
| T1 model | LR (TD/VD) | 1.000/0.643 | 1.000/0.636 | 1.000/0.700 | 1.000/0.609 | 1.000/0.438 |
| RF (TD/VD) | 0.949/0.687 | 0.823/0.697 | 0.395/0.100 | 1.000/0.957 | 1.000/0.500 | |
| SVM (TD/VD) | 1.000/0.513 | 1.000/0.697 | 1.000/0.000 | 1.000/1.000 | 1.000/0.000 | |
| T1A model | LR (TD/VD) | 1.000/0.443 | 1.000/0.636 | 1.000/0.500 | 1.000/0.304 | 1.000/0.238 |
| RF (TD/VD) | 0.967/0.700 | 1.000/0.364 | 0.158/0.000 | 1.000/1.000 | 1.000/0.000 | |
| SVM (TD/VD) | 1.000/0.500 | 0.754/0.697 | 1.000/0.000 | 1.000/1.000 | 1.000/0.000 | |
| T1D model | LR (TD/VD) | 1.000/0.665 | 1.000/0.606 | 1.000/0.700 | 1.000/0.565 | 1.000/0.412 |
| RF (TD/VD) | 0.978/0.765 | 0.738/0.697 | 0.105/0.000 | 1.000/1.000 | 1.000/0.000 | |
| SVM (TD/VD) | 1.000/0.574 | 1.000/0.697 | 1.000/0.000 | 1.000/1.000 | 1.000/0.000 | |
| T1V model | LR (TD/VD) | 1.000/0.430 | 1.000/0.424 | 1.000/0.600 | 1.000/0.348 | 1.000/0.286 |
| RF (TD/VD) | 0.979/0.661 | 0.738/0.697 | 0.105/0.000 | 1.000/1.000 | 1.000/0.000 | |
| SVM (TD/VD) | 1.000/0.426 | 1.000/0.697 | 1.000/0.000 | 1.000/1.000 | 1.000/0.000 | |
| T2 model | LR (TD/VD) | 1.000/0.422 | 1.000/0.424 | 1.000/0.100 | 1.000/0.565 | 1.000/0.091 |
| RF (TD/VD) | 0.969/0.383 | 0.746/0.697 | 0.132/0.000 | 1.000/1.000 | 1.000/0.000 | |
| SVM (TD/VD) | 1.000/0.448 | 1.000/0.697 | 1.000/0.000 | 1.000/1.000 | 1.000/0.000 | |
| DWI+T1 model | LR (TD/VD) | 0.941/0.817 | 0.892/0.758 | 0.895/0.800 | 0.891/0.739 | 0.773/0.571 |
| RF (TD/VD) | 0.963/0.854 | 0.908/0.848 | 0.895/0.900 | 0.913/0.826 | 0.810/0.692 | |
| SVM (TD/VD) | 0.941/0.826 | 0.892/0.788 | 0.816/0.800 | 0.924/0.783 | 0.816/0.615 | |
| DWI+T1D model | LR (TD/VD) | 0.901/0.852 | 0.846/0.788 | 0.684/0.700 | 0.913/0.826 | 0.765/0.636 |
| RF (TD/VD) | 0.897/0.852 | 0.792/0.636 | 0.816/0.800 | 0.783/0.565 | 0.608/0.444 | |
| SVM (TD/VD) | 0.890/0.835 | 0.815/0.788 | 0.474/0.600 | 0.957/0.870 | 0.818/0.667 | |
| T1+T1D model | LR (TD/VD) | 0.883/0.874 | 0.846/0.818 | 0.711/0.600 | 0.902/0.913 | 0.705/0.750 |
| RF (TD/VD) | 0.905/0.878 | 0.869/0.818 | 0.816/0.800 | 0.891/0.826 | 0.756/0.667 | |
| SVM (TD/VD) | 0.884/0.835 | 0.777/0.727 | 0.237/0.100 | 1.000/1.000 | 1.000/1.000 | |
| Radiomics model | LR (TD/VD) | 0.950/0.883 | 0.862/0.788 | 0.921/0.900 | 0.837/0.739 | 0.700/0.600 |
| RF (TD/VD) | 0.967/0.891 | 0.908/0.879 | 0.895/0.900 | 0.913/0.870 | 0.801/0.750 | |
| SVM (TD/VD) | 0.947/0.865 | 0.869/0.818 | 0.579/0.700 | 0.989/0.870 | 0.957/0.700 | |
| Imaging+radiomics model | LR (TD/VD) | 0.953/0.861 | 0.892/0.818 | 0.974/0.900 | 0.859/0.783 | 0.740/0.643 |
| RF (TD/VD) | 0.988/0.878 | 0.946/0.909 | 0.895/0.800 | 0.967/0.957 | 0.919/0.889 | |
| SVM (TD/VD) | 0.898/0.878 | 0.869/0.909 | 0.763/0.900 | 0.913/0.913 | 0.784/0.818 | |
| Radiomics model | LR (test) | 0.812 (0.617–1.000) | 0.792 | 0.750 | 0.833 | 0.818 |
| RF (test) | 0.757 (0.532–0.982) | 0.792 | 0.667 | 0.917 | 0.889 | |
| SVM (test) | 0.812 (0.617–1.000) | 0.708 | 0.500 | 0.917 | 0.857 | |
| Imaging+radiomics model | LR (test) | 0.819 (0.620–1.000) | 0.875 | 0.833 | 0.917 | 0.909 |
| RF (test) | 0.771 (0.556–0.986) | 0.750 | 0.583 | 0.917 | 0.875 | |
| SVM (test) | 0.771 (0.555–0.987) | 0.792 | 0.667 | 0.917 | 0.889 |
LR, logistic regression; RF, random forest; SVM, support vector machine; TD, training dataset; VD, validation dataset; AUC, area under the curve.