Table 4 Comparison of our work with state-of-the-art methods developed for automated pain intensity classification using facial images.
Study | Method | Classifier | Dataset | Results |
|---|---|---|---|---|
Bargshady35 | Temporal convolutional network, LSTM, principal component analysis | Temporal convolutional network | UNBC-McMaster (10,783 frames) | MSE: 1.186 |
MAE: 0.234 | ||||
Acc: 94.14% | ||||
AUC: 91.30% | ||||
Bargshady36 | Ensemble neural network | Ensemble CNN-recurrent neural network | UNBC-McMaster (10,783 frames) | AUC: 90.50% |
Acc: 86.00% | ||||
MSE: 0.081 | ||||
Semwal37 | Ensemble of compact CNN | Ensemble | UNBC-McMaster (16,000 frames) | Pre: 91.97% |
Rec: 91.01$ | ||||
F1: 91.42% | ||||
Acc: 93.87 | ||||
Rudovic38 | CNN | Softmax | UNBC-McMaster (48,106 frames) | Acc: 76.00% |
PR-AUC: 59.00 | ||||
F1: 47.00 | ||||
Karamitsos39 | CNN | Softmax | UNBC-McMaster (48,398 frames) | Acc: 92.50% |
Semwal40 | CNN | Softmax | UNBC-McMaster (16,000 frames) | Acc: 92.00% |
MAE: 0.20 | ||||
MSE: 0.17 | ||||
Bargshady1 | CNN, bidirectional LSTM | Enhanced joint hybrid-CNN-bidirectional LSTM | UNBC-McMaster (10,783 frames) | Acc: 91.20% |
AUC: 98.40% | ||||
El Morabit and Rivenq41 | Vision Transformer, Feed Forward Network | Softmax | UNBC-McMaster (48,398 frames) | Acc: 84.15 |
Our model | Transfer learning, novel shutter blinds-based deep feature extraction | kNN | UNBC-McMaster (10,852 frames) | Acc: 95.57% |
UAR: 95.59% | ||||
UAP: 95.79% | ||||
Average F1: 95.67% | ||||
MCC: 94.14% | ||||
CK: 93.93% | ||||
GM: 95.58% | ||||
DISFA (39,182 frames) | Acc: 96.06% | |||
UAR: 96.04% | ||||
UAP: 96.16% | ||||
Average F1: 96.08% | ||||
MCC: 94.78% | ||||
CK: 94.74% | ||||
GM: 96.03% |
