Table 2 Performance comparison of different modal features and different fusion methods
From: Hybrid-FHR: a multi-modal AI approach for automated fetal acidosis diagnosis
Method | Acc% | Spe% | Sen% | Pre% | F1% |
|---|---|---|---|---|---|
Morphological time domain features only | 79.3 | 76 | 82.5 | 77.5 | 79.9 |
Frequency domain features only | 80.8 | 77.5 | 84 | 78.9 | 81.4 |
Nonlinear features only | 75.5 | 80.5 | 70.5 | 78.3 | 74.2 |
All expert features | 86.8 | 83.5 | 90.0 | 84.5 | 87.1 |
Signal features only | 92.0 | 95.0 | 89.0 | 94.7 | 91.8 |
Early fusion (concatenation fusion) | 93.5 | 97.0 | 90.0 | 96.8 | 93.3 |
Late fusion (decision-level fusion) | 95.8 | 96.0 | 95.5 | 96.0 | 95.7 |
Hybrid-FHR (Expert features + Signal features + CMFF) | 96.8 | 97.5 | 96.0 | 97.5 | 96.7 |