Table 1. List of pipeline steps, and choices tested for each step.
Steps that are varied during each subject’s pipeline optimization are in bold, and other steps are held fixed. We tested pipeline optimization for fixed analysis models: Gaussian Naïve Bayes (GNB; univariate) and Canonical Variates Analysis (CVA; multivariate). CVA analysis is performed using 1 to k Principal Components (PCs), where we vary k = 1 to 10.
| PIPELINE STEPS | CHOICES |
|---|---|
| 1. Estimate minimum-displacement brain volume | ON |
| 2. Rigid-body motion correction | OFF / ON |
| 3. Censoring of outlier brain volumes | OFF / ON |
| 4. Physiological correction; external physiological measures (RETROICOR) | OFF / ON |
| 5. Slice-timing correction | OFF / ON |
| 6. Spatial smoothing | 6mm FWHM |
| 7. Subject-specific non-neuronal tissue mask | ON |
| 8. Temporal detrending | orders 0 to 5 |
| 9. Motion parameter regression | OFF / ON |
| 10. Global signal regression using Principal Component Analysis (PCA) | OFF / ON |
| 11. Including task design as a regressor | OFF / ON |
| 12. Physiological correction; multivariate data-driven model (PHYCAA+) | OFF / ON |
| 13. Analysis model: univariate (GNB) or multivariate (CVA)* | GNB or CVA |