Table 3 The predictive performance results of radiomics machine learning signatures without SMOTE
Set | Classifier | AUC (95%CI) | ACC | SEN | SPE | PPV | NPV |
|---|---|---|---|---|---|---|---|
Training | SVM | 0.932 (0.889–0.975) | 0.795 | 0 | 1.000 | 0 | 0.795 |
LR | 0.749 (0.660–0.839) | 0.811 | 0.105 | 0.993 | 0.800 | 0.811 | |
XGBoost | 1 (1.000–1.000) | 0.995 | 0.974 | 1.000 | 1.000 | 0.993 | |
NaiveBayes | 0.720 (0.631–0.809) | 0.373 | 1.000 | 0.211 | 0.247 | 1 | |
AdaBoost | 0.879 (0.828–0.930) | 0.854 | 0.447 | 0.959 | 0.739 | 0.870 | |
LightGBM | 0.932 (0.888–0.977) | 0.805 | 0.053 | 1.000 | 1.000 | 0.803 | |
KNN | 0.781 (0.715–0.848) | 0.805 | 0.237 | 0.952 | 0.563 | 0.828 | |
MLP | 0.745 (0.657–0.833) | 0.795 | 0 | 1.000 | 0 | 0.795 | |
GradientBoosting | 0.945 (0.895–0.996) | 0.843 | 0.237 | 1.000 | 1.000 | 0.835 | |
External test | SVM | 0.659 (0.465–0.853) | 0.815 | 0 | 1.000 | 0 | 0.815 |
LR | 0.689 (0.480–0.897) | 0.815 | 0 | 1.000 | 0 | 0.815 | |
XGBoost | 0.720 (0.524–0.917) | 0.815 | 0 | 1.000 | 0 | 0.815 | |
NaiveBayes | 0.764 (0.604–0.924) | 0.704 | 0.700 | 0.705 | 0.350 | 0.912 | |
AdaBoost | 0.699 (0.510–0.888) | 0.815 | 0.100 | 0.977 | 0.500 | 0.827 | |
LightGBM | 0.639 (0.447–0.830) | 0.815 | 0 | 1.000 | 0 | 0.815 | |
KNN | 0.682 (0.531–0.833) | 0.796 | 0 | 0.977 | 0 | 0.811 | |
MLP | 0.727 (0.570–0.884) | 0.815 | 0 | 1.000 | 0 | 0.815 | |
GradientBoosting | 0.616 (0.385–0.847) | 0.815 | 0 | 1.000 | 0 | 0.815 |