Table 2.
Performance on classification between individuals with schizophrenia and controls
| GCN | SVM | |||||
|---|---|---|---|---|---|---|
| Model performance | BAC (%) | SEN (%) | SPE (%) | BAC (%) | SEN (%) | SPE (%) |
| Dataset 1 | 68.8 | 74.5 | 63.1 | 73.4 | 80.1 | 66.8 |
| Dataset 2 | - | - | - | - | - | - |
| Dataset 3 | 79.2 | 78.2 | 80.2 | 67.2 | 63.0 | 71.4 |
| Dataset 4 | 79.0 | 74.3 | 83.7 | 75.6 | 55.8 | 95.4 |
| Dataset 5 | 72.3 | 73.7 | 71.0 | 73.8 | 70.0 | 77.7 |
| Dataset 6 | 65.7 | 52.7 | 78.6 | 72.5 | 53.3 | 91.7 |
| LOSO (Before ComBat) | 61.0 | 60.4 | 61.6 | 62.3 | 56.4 | 68.2 |
| LOSO (After ComBat) | 79.1 | 85.0 | 73.2 | 73.4 | 60.6 | 86.2 |
| 10-fold | 85.8 | 74.0 | 97.6 | 80.9 | 69.9 | 91.9 |
Abbreviations: GCN, graph convolutional network; SVM, support vector machine; BAC, balanced accuracy; SEN, sensitivity; SPE, specificity; LOSO, leave-one-site-out.