Table 9 Comparison of studies using the AI4PAIN database for three-class pain (No Pain, Low Pain, High Pain) classification settings.
Reference | Modality | Validation | Method | Accuracy (%) |
|---|---|---|---|---|
Gkikas et al.52 | Video, fNIRS | Hold-out | Transformer | 46.67 |
Prajod et al.53 | Video | Hold-out | 2D CNN | 49.00 |
Khan et al.54 | fNIRS | Hold-out | Ensemble Learning | 53.66 |
Nguyen et al.55 | Video | Hold-out | Transformer | 55.00 |
Vianto et al.56 | Video, fNIRS | Hold-out | Hybrid CNN–Transformer | 51.33 |
Gkikas et al.47 | Video, fNIRS | Hold-out | Transformer | 55.69 |
Bargshady et al.9 | Video | Hold-out | Transformer | 66.96 |
Fernandez-Rojas et al.35 | EDA | Hold-out | Gaussian SVM | 45.60 |
Fernandez-Rojas et al.35 | BVP | Hold-out | Gaussian SVM | 56.25 |
Fernandez-Rojas et al.35 | EDA, BVP | Hold-out | Gaussian SVM | 52.77 |
Our Approach | EDA, BVP | Hold-out | Crossmod-Transformer | 75.83 |
