Table 4.
Comparison of classification performance of various feature combinations (%).
| Feature | Model | Macro-AUC | ACC | Macro-R | Macro-P | Macro-F1 |
|---|---|---|---|---|---|---|
| Clinical | EDL | 97.15 | 88.30 | 87.60 | 86.78 | 86.82 |
| Radiomics | EDL | 90.79 | 90.74 | 74.10 | 80.28 | 75.82 |
| Joint | EDL | 96.68 | 92.55 | 92.10 | 91.42 | 91.72 |
| Clinical | OEDL | 96.13 | 90.43 | 90.57 | 89.29 | 89.35 |
| Radiomics | OEDL | 90.50 | 93.21 | 82.19 | 86.27 | 83.87 |
| Joint | OEDL | 97.89 | 95.74 | 94.75 | 94.03 | 94.35 |
EDL represents a deep ensemble learning model based on DNN, LSTM-RNN, DBN, and stacking ensemble; OEDL is an optimization algorithm based on EDL and BBOA.
These bold characters represent the predictive performance of the optimal method.