Table 5. Summary of classifier performance.
Model | Testing Dataset | Sensitivity[95% CI] | Specificity[95% CI] | Best CVA[95% CI] | Noise/Not Noise |
M1 | SchizophreniaSame institution, same scanner | 0.890.86 0.93 | 0.830.78 0.87 | 0.860.83 0.88 | 343 (295) |
M1 | NKI Rockland InstituteDifferent institution, same scanner | 0.880.86 0.90 | 0.880.86 0.91 | 0.880.86 0.89 | 947 (711) |
M1 | Human Connectome DatabaseDifferent institution, different scanner | 0.720.68 0.77 | 0.920.88 0.95 | 0.790.75 0.82 | 451 (230) |
Performance metrics (sensitivity, specificity, and best cross validation accuracy (CVA)) and proportion of noise components in data for model of all comprehensive noise (All Noise, M1) built with Data A and tested with ten -fold cross validation on three novel datasets: Data B (same institution, same scanner, different subject population), Data C (different institution, same scanner), Data D (different institution, different scanner).