Table 4.
1p/19q status classification performance in LGGs (WHO 2016).
| Classification Metrics | cMRI | Multi-Shell dMRI | cMRI and Multi-Shell dMRI |
|---|---|---|---|
| Overall performance | |||
| Accuracy | 65% ± 6% | 66% ± 9% | 72% ± 4% |
| Precision 1 | 0.66 ± 0.06 | 0.73 ± 0.10 | 0.75 ± 0.05 |
| MCC | 0.30 ± 0.10 | 0.38 ± 0.20 | 0.45 ± 0.08 |
| Individual group sensitivity | |||
| 1p/19q codeleted | 70% ± 10% | 60% ± 10% | 60% ± 10% |
| 1p/19q uncodeleted | 61% ± 20% | 70% ± 20% | 83% ± 10% |
Classification performance of ResNet on the unseen test sets with different MRI inputs to predict 1p/19q codeletion versus 1p/19q non-codeletion (IDH-mutant or IDH-wildtype) in LGGs according to the 2016 WHO classification. Data are reported as average ± standard deviation over the five test sets in the outer loop. Overall, data of 60% (60 out of 100) LGGs patients were tested as unseen set. 1 Computed as the macro-average of the precisions for the IDH-mutant group and the IDH-wildtype group. LGG = lower-grades glioma; cMRI = conventional MRI; dMRI = diffusion MRI; MCC = Matthews correlation coefficient.