Table 1.
The scanning parameters of ABIDE dataset for each site show the different in MRI Scanner, TR (Repetition Time), TE (Echo Time), Flip Angle, and Age which may result in difference in data acquisition as well as the pre- and post-processing of fMRI data.
| Site | MRI scanner | TR (ms) | TE (ms) | Flip angle (degree) | Age (year) |
|---|---|---|---|---|---|
| Caltech | SIEMENS | 2,000 | 30 | 75 | 17–56.2 |
| CMU | SIEMENS | 2,000 | 30 | 73 | 19–40 |
| KKI | PHILLIPS | 2,500 | 30 | 75 | 8–12.8 |
| Leuven | PHILLIPS | 1,656 | 33 | 90 | 12.1–32 |
| MaxMun | SIEMENS | 3,000 | 30 | 80 | 7–58 |
| NYU | SIEMENS | 2,000 | 15 | 90 | 6.5–39.1 |
| OHSU | SIEMENS | 2,500 | 30 | 90 | 8–15.2 |
| OLIN | SIEMENS | 1,500 | 27 | 60 | 10–24 |
| PITT | SIEMENS | 1,500 | 25 | 70 | 9.3–35.2 |
| SBL | PHILLIPS | 2,200 | 30 | 80 | 20–64 |
| SDSU | GE | 2,000 | 30 | 90 | 8.7–17.2 |
| Stanford | GE | 2,000 | 30 | 80 | 7.5–12.9 |
| Trinity | PHILLIPS | 2,000 | 28 | 90 | 12–25.9 |
| UCLA | SIEMENS | 3,000 | 28 | 90 | 8.4–17.9 |
| UM | GE | 2,000 | 30 | 90 | 8.2–28.8 |
| USM | SIEMENS | 2,000 | 28 | 90 | 8.8–50.2 |
| Yale | SIEMENS | 2,000 | 25 | 60 | 7–17.8 |
These differences may lead the machine-learning models learn site-specific variations leading many machine-learning models give better average accuracy (for whole ABIDE data set) than the site-specific accuracy.