Table 2 Classification performance metrics for the five datasets using nine classifiers
Classifier | Dataset Name | AUC | Balanced Accuracy |
|---|---|---|---|
SVM | just the radiomics features | 0.91 ± 0.013 | 0.76 ± 0.060 |
just the deep features | 0.94 ± 0.021 | 0.77 ± 0.065 | |
PCA + deep features | 0.94 ± 0.023 | 0.79 ± 0.048 | |
PCA + radiomics features | 0.92 ± 0.030 | 0.76 ± 0.060 | |
combination of deep features and radiomics features | 0.94 ± 0.014 | 0.77 ± 0.045 | |
Logistic Regression | just the radiomics features | 0.89 ± 0.023 | 0.70 ± 0.057 |
just the deep features | 0.93 ± 0.028 | 0.76 ± 0.064 | |
PCA + deep features | 0.88 ± 0.020 | 0.72 ± 0.052 | |
PCA + radiomics features | 0.77 ± 0.029 | 0.61 ± 0.060 | |
combination of deep features and radiomics features | 0.93 ± 0.016 | 0.77 ± 0.045 | |
Gaussian Naive Bayes | just the radiomics features | 0.91 ± 0.028 | 0.67 ± 0.074 |
just the deep features | 0.90 ± 0.027 | 0.67 ± 0.057 | |
PCA + deep features | 0.88 ± 0.055 | 0.67 ± 0.107 | |
PCA + radiomics features | 0.72 ± 0.046 | 0.39 ± 0.091 | |
combination of deep features and radiomics features | 0.92 ± 0.018 | 0.69 ± 0.046 | |
KNN | just the radiomics features | 0.80 ± 0.043 | 0.66 ± 0.081 |
just the deep features | 0.77 ± 0.023 | 0.64 ± 0.037 | |
PCA + deep features | 0.79 ± 0.025 | 0.65 ± 0.048 | |
PCA + radiomics features | 0.89 ± 0.040 | 0.70 ± 0.054 | |
combination of deep features and radiomics features | 0.79 ± 0.20 | 0.65 ± 0.029 | |
Random Forest | just the radiomics features | 0.93 ± 0.017 | 0.74 ± 0.043 |
just the deep features | 0.94 ± 0.028 | 0.78 ± 0.078 | |
PCA + deep features | 0.92 ± 0.029 | 0.72 ± 0.081 | |
PCA + radiomics features | 0.88 ± 0.028 | 0.68 ± 0.043 | |
combination of deep features and radiomics features | 0.94 ± 0.031 | 0.76 ± 0.086 | |
Bagging + Decision Tree | just the radiomics features | 0.92 ± 0.016 | 0.70 ± 0.036 |
just the deep features | 0.93 ± 0.014 | 0.74 ± 0.048 | |
PCA + deep features | 0.90 ± 0.036 | 0.69 ± 0.098 | |
PCA + radiomics features | 0.84 ± 0.024 | 0.61 ± 0.058 | |
combination of deep features and radiomics features | 0.93 ± 0.029 | 0.73 ± 0.051 | |
Gradient Boosting | just the radiomics features | 0.93 ± 0.019 | 0.73 ± 0.033 |
just the deep features | 0.94 ± 0.026 | 0.76 ± 0.082 | |
PCA + deep features | 0.92 ± 0.030 | 0.74 ± 0.048 | |
PCA + radiomics features | 0.88 ± 0.045 | 0.68 ± 0.076 | |
combination of deep features and radiomics features | 0.94 ± 0.021 | 0.78 ± 0.046 | |
Ensemble classifier (Voting 1) | just the radiomics features | 0.92 ± 0.023 | 0.74 ± 0.033 |
just the deep features | 0.93 ± 0.021 | 0.78 ± 0.045 | |
PCA + deep features | 0.92 ± 0.020 | 0.74 ± 0.067 | |
PCA + radiomics features | 0.86 ± 0.031 | 0.60 ± 0.065 | |
combination of deep features and radiomics features | 0.94 ± 0.024 | 0.77 ± 0.055 | |
Ensemble classifier (Voting 2) | just the radiomics features | 0.92 ± 0.023 | 0.74 ± 0.012 |
just the deep features | 0.95 ± 0.020 | 0.78 ± 0.065 | |
PCA + deep features | 0.94 ± 0.024 | 0.76 ± 0.068 | |
PCA + radiomics features | 0.90 ± 0.022 | 0.69 ± 0.033 | |
combination of deep features and radiomics features | 0.94 ± 0.018 | 0.77 ± 0.033 |