Table 3 Classification performance of features combination models.
From: Automated classification of chondroid tumor using 3D U-Net and radiomics with deep features
Classifiers | Accuracy (95% CI) | Weighted Kappa | AUC |
|---|---|---|---|
Model 1: radiomics-only features model | |||
Random forest | 0.75 (0.75–0.82) | 0.57 | 0.75 |
Gradient boosting | 0.68 (0.68–0.75) | 0.42 | 0.68 |
XGBoost | 0.75 (0.75–0.82) | 0.62 | 0.75 |
LightGBM | 0.69 (0.68–0.75) | 0.42 | 0.68 |
CatBoost | 0.81 (0.81–0.87) | 0.69 | 0.81 |
Model 2: deep learning-only features model | |||
Random forest | 0.74 (0.74–0.80) | 0.52 | 0.63 |
Gradient boosting | 0.78 (0.78–0.84) | 0.61 | 0.74 |
XGBoost | 0.78 (0.78–0.84) | 0.58 | 0.74 |
LightGBM | 0.81 (0.81–0.87) | 0.65 | 0.77 |
CatBoost | 0.84 (0.84–0.89) | 0.71 | 0.79 |
Model 3: radiomics + deep learning model | |||
Random forest | 0.81 (0.81–0.86) | 0.76 | 0.71 |
Gradient boosting | 0.84 (0.84–0.89) | 0.82 | 0.80 |
XGBoost | 0.87 (0.87–0.91) | 0.84 | 0.83 |
LightGBM | 0.83 (0.83–0.87) | 0.79 | 0.85 |
CatBoost | 0.87 (0.86–0.92) | 0.84 | 0.85 |
Model 4: radiomics + deep learning + clinical model | |||
Random forest | 0.80 (0.80–0.84) | 0.64 | 0.73 |
Gradient boosting | 0.83 (0.83–0.86) | 0.70 | 0.85 |
XGBoost | 0.80 (0.80–0.84) | 0.66 | 0.83 |
LightGBM | 0.81 (0.81–0.85) | 0.69 | 0.86 |
CatBoost | 0.89 (0.88–0.92) | 0.78 | 0.87 |
Model 5: Radiomics + deep roi + clinical model | |||
Random forest | 0.88 (0.87–0.90) | 0.77 | 0.75 |
Gradient boosting | 0.89 (0.88–0.91) | 0.80 | 0.88 |
XGBoost | 0.89 (0.88–0.92) | 0.81 | 0.89 |
LightGBM | 0.90 (0.89–0.92) | 0.80 | 0.90 |
CatBoost | 0.90 (0.90–0.93) | 0.85 | 0.91 |
