From: Generative adversarial networks in EEG analysis: an overview
Study | Purpose | Dataset | GAN type | Evaluation metrics | Results (with GAN) |
---|---|---|---|---|---|
Wei et al., 2019 [67] | Proposes an automatic epileptic EEG detection method | CHB-MIT Scalp | WGANs | • Classification accuracy • Sensitivity • Specificity | • + 2.51% • + 1.43% • + 3.59% |
You et al., 2020 [68] | Solve the class imbalance problem of epileptic seizures detection | Private dataset | DCGAN an anomaly detector | • AUROC • Sensitivity • False detection rate | • 93.93 1% (With Gram Matrix) • 96.3% • 0.14 per hour |
Pascual et al., 2021 [70] | Overcome scarcity of epileptic seizures EEG signals and address the privacy concerns | EPILEPSIAE project [73] | Epilepsy GAN | • Classification accuracy • Synthetic data Recall values • Geometric mean of sensitivity and specificity | • + 1.3% • median: + 3.2% • + 1.3% |
Truong et al., 2019 [74] | Predict seizures with an unsupervised algorithm | CHB-MIT + Freiburg Hospital + EPILEPSIAE | DCGAN feature extractor | • Classification accuracy | CHB-MIT: 61.53% Freiburg Hospital: 53.84% EPILEPSIAE 13.33% (with AUC above 80%) |
Usman et al. 2021 [71] | Solve the class imbalance problem of epileptic seizures predictor | CHBMIT | GAN | • Sensitivity • Specificity • Anticipation time | • 93% • 92.5% • Average 32 min |
Usman et al., 2021 [75] | Overcome the challenge of accurate prediction of epileptic seizures | CHBMIT + American epilepsy society-Kaggle seizure prediction | GAN | Classification accuracy | • CHB-MIT: + 1.74% • IEEG: achieved 95.53% |
Sensitivity | • CHB-MIT: + 1.56% • IEEG: 94.27% | ||||
Specificity | • CHB-MIT: + 1.93% • IEEG: 95.81% | ||||
Average Anticipation Time | • CHB-MIT: 1.34 m | ||||
Salazar et al., 2021 [76] | Improve seizure prediction performance with extreme data scarcity | Private dataset “Barcelona test” | GAN + vector Markov Random Field (vMRF) | Classification accuracy | NA |
Rasheed et al. 2021 [77] | Improve seizure prediction performance | Epilepsyecosystem + CHB-MIT | DCGAN | • Sensitivity • AUC | • + 15% • + 6:10% |