Table 4.
Comparison between CNN-based predictions and radiomics-based predictions with different inputs (Mean ± STD).
| Method | Input | ACC (%) | SEN (%) | SPE (%) | AUC |
|---|---|---|---|---|---|
| T1 + annotation | 75.76 ± 2.02 | 74.47 ± 3.05 | 76.70 ± 3.74 | 0.8293 ± 0.0221 | |
| CNN | CEST + annotation | 74.88 ± 2.58 | 74.37 ± 4.24 | 75.28 ± 2.27 | 0.8192 ± 0.0216 |
| CEST + T1 + annotation | 82.94 ± 1.23 | 82.34 ± 1.87 | 83.45 ± 2.50 | 0.8868 ± 0.0055 | |
| Rad +LR +L1 |
T1 + annotation | 67.36 ± 1.70*** | 67.16 ± 2.35** | 67.57 ± 1.81** | 0.6736 ± 0.0170*** |
| CEST + annotation | 72.03 ± 1.37* | 77.51 ± 3.82 | 66.89 ± 2.68** | 0.7220 ± 0.0151*** | |
| CEST + T1 + annotation | 76.35 ± 1.98** | 77.95 ± 3.04* | 74.89 ± 1.92** | 0.7641 ± 0.0204*** | |
| Rad +SVM +RBF |
T1 + annotation | 54.80 ± 1.23*** | 68.83 ± 4.04*** | 41.92 ± 1.53*** | 0.5537 ± 0.0156*** |
| CEST + annotation | 58.08 ± 10.1*** | 83.30 ± 3.00*** | 34.68 ± 2.6*** | 0.5899 ± 0.0082*** | |
| CEST + T1 + annotation | 64.13 ± 1.91*** | 66.30 ± 2.61*** | 62.11 ± 1.73*** | 0.6420 ± 0.0194*** |
*DL-based model is significantly better than the radiomics-based method using same inputs at p < 0.05 level. **]p < 0.01. ***]p < 0.001. Rad, Radiomics; LR, logistic regression; SVM, support vector machine; L1, L1-penalty; RBF, radial basis function kernel.