Table 2.
Optimization of the best classifier in multilayer perceptron network (metric 2) and CNN (metric 4) by comparing diagnostic performance in the internal validation set.
| Imaging data | CE-T1WI + ADC | CE-T1WI | ADC |
|---|---|---|---|
| MLP network | AUC (95% CI) | ||
| 100-10 | 0.991 (0.987–0.994) | 0.965 (0.959–0.972) | 0.969 (0.960–0.978) |
| 500-100-10 | 0.990 (0.987–0.993) | 0.965 (0.960–0.971) | 0.968 (0.956–0.979) |
| 500-100-50-10 | 0.989 (0.985–0.993) | 0.956 (0.949–0.962) | 0.964 (0.953–0.975) |
| 500-250-100-50-10 | 0.988 (0.982–0.995) | 0.968 (0.964–0.982) | 0.965 (0.952–0.977) |
| 750-500-250-100-50-10 | 0.986 (0.978–0.993) | 0.971 (0.967–0.975) | 0.960 (0.951–0.969) |
| CNN | Accuracy (Sensitivity/Specificity) | ||
| Inception v-3 | 80.0 (50.0/100) | 76.7 (41.6/100) | 86.7 (83.3/88.9) |
Abbreviations: MLP = multilayer perceptron, CNN = convolutional neural network, CE-T1WI = contrast-enhanced T1 weighted imaging, DWI = diffusion weighted imaging, AUC = area under the receiver operating characteristic curve.