Table 4.
Ex-vivo classification results from 10-fold cross validation using the top six experimentally accessible response features as inputs to the C-SBINN.
| Top 6 features selected from simulated data, with transfer learning | ||||||
| ROC (AUC) | PRC (AUC) | MCC | CKS | MCC (all) | CKS (all) | Optimal threshold |
| 0.82 ± 0.16 | 0.62 ± 0.28 | 0.72 ± 0.21 | 0.68 ± 0.22 | 0.63 ± 0.12 | 0.61 ± 0.13 | 0.28 ± 0.12 |
| Top 6 features selected from simulated data, without transfer learning | ||||||
| ROC (AUC) | PRC (AUC) | MCC | CKS | MCC (all) | CKS (all) | Optimal threshold |
| 0.75 ± 0.19 | 0.52 ± 0.27 | 0.62 ± 0.22 | 0.56 ± 0.25 | 0.63 ± 0.20 | 0.59 ± 0.23 | 0.31 ± 0.28 |
| Top 6 features selected from ex-vivo data, without transfer learning | ||||||
| ROC (AUC) | PRC (AUC) | MCC | CKS | MCC (all) | CKS (all) | Optimal threshold |
| 0.71 ± 0.20 | 0.59 ± 0.27 | 0.59 ± 0.25 | 0.55 ± 0.28 | 0.39 ± 0.16 | 0.36 ± 0.16 | 0.78 ± 0.33 |
| All ex-vivo experimental measurements, without transfer learning | ||||||
| ROC (AUC) | PRC (AUC) | MCC | CKS | MCC (all) | CKS (all) | Optimal threshold |
| 0.75 ± 0.13 | 0.39 ± 0.20 | 0.59 ± 0.17 | 0.52 ± 0.21 | 0.64 ± 0.13 | 0.60 ± 0.17 | 0.25 ± 0.23 |
Average and standard deviation are shown. First two rows correspond to response features selected from the simulated clinical trial data. Third and fourth rows correspond to, respectively, the top six experimental features selected directly from the ex-vivo data and all the experimental ex-vivo measurements. MCC (all) and CKS (all) refer to an additional validation step wherein the trained C-SBINN was applied to the entire ex-vivo data set after optimizing the classification threshold on the testing set only.