Table 4:
Generality comparisons among three algorithms across datasets. Here, the generality as measured by the MAEs in each dataset in the discovery and replication data, as mentioned in Section 5.3(a).
| ResNetfus (He et al., 2016) |
Hi-Netfus (Zhou et al., 2020) |
FiA-Netfus |
|||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Dataset | Nsubjects | Age range (years) | MAE (years) | CS (α=2 years) | CS (α=5 years) | MAE | CS (α=2 years) | CS(α=5 years) | MAE | CS (α=2 years) | CS (α=5 years) | ||||
| Discovery Data | BeijingEN | 180 | 17–28 | 2.14 | 59.4% | 91.7% | 1.86 | 68.3% | 95.0% | 1.73 | 65.0% | 96.7% | |||
| DLBS | 315 | 20–89 | 5.95 | 22.5% | 50.8% | 6.29 | 20.3% | 47.9% | 5.58 | 23.8% | 54.6% | ||||
| MGHBCH | 428 | 0–6 | 1.81 | 72.7% | 92.8% | 0.90 | 91.8% | 98.4% | 0.99 | 90.0% | 98.4% | ||||
| IXI | 556 | 20–86 | 6.10 | 24.3% | 51.4% | 6.28 | 22.3% | 47.7% | 6.3 | 20.5% | 46.3% | ||||
| ABIDE-I | 567 | 6–56 | 3.51 | 42.0% | 79.9% | 3.22 | 49.0% | 82.2% | 3.07 | 49.9% | 84.6% | ||||
| NIH-PD | 1,211 | 0–22 | 1.59 | 69.3% | 96.6% | 1.44 | 75.1% | 97.6% | 1.31 | 80.9% | 97.8% | ||||
| BGSP | 1,570 | 19–53 | 2.03 | 64.3% | 91.4% | 1.93 | 65.9% | 93.5% | 1.88 | 66.8% | 93.7% | ||||
| OASIS-3 | 1,222 | 42–97 | 4.70 | 29.9% | 62.3% | 4.42 | 32.2% | 67.0% | 4.22 | 35.0% | 70.4% | ||||
| All | 6,049 | 0–97 | 3.31 | 49.1% | 49.4% | 3.12 | 53.0% | 80.5% | 3.00 | 54.8% | 82.1% | ||||
| Replication Datal | ABCD | 8,639 | 9–12 | 3.17 | 34.8% | 81.7% | 3.55 | 37.0% | 76.4% | 2.86 | 48.8% | 92.3% | |||
| CMI | 1,765 | 5–22 | 3.29 | 38.7% | 80.4% | 3.79 | 37.4% | 74.1% | 2.97 | 44.1% | 83.8% | ||||
| CoRR | 252 | 6–60 | 6.43 | 32.6% | 60.6% | 5.47 | 37.5% | 66.7% | 5.35 | 41.0% | 66.6% | ||||
| All | 10,656 | 5–60 | 4.29 | 35.4% | 74.3% | 4.27 | 37.3% | 72.4% | 3.72 | 44.6% | 80.9% | ||||