Table 1.
The summary of existing works using 1-D and 2-D ECG input.
| Ref | Dataset | Signal used | ECG input | Purposes | Feature extracted | Classifier | Result (%) |
|---|---|---|---|---|---|---|---|
| 21 | Own dataset
69 subjects, 20 stimuli |
ECG | 1-D | ERS | Statistical features from the time and frequency domains | SVM, NB, KNN, Gaussian | SVM – 69.23
NB – 53.83 KNN – 61.83 Gaussian – 70.00 |
| 22 | AUBT | ECG | 1-D | ERS | Local pattern description using Local Binary Pattern (LBP) and Local Ternary Pattern (LTP) | KNN | LBP – 84.17
LTP – 87.92 |
| 23 | (MPED)
23 subjects, 28 stimuli |
ECG | 1-D | ERS | Statistical features from the time and frequency domains | SVM, KNN, LSTM, A-LSTM | SVM – 42.66
KNN – 40.02 LSTM – 49.77 A-LSTM – 51.66 |
| 13 | (DREAMER)
23 subject, 18 stimuli |
ECG, EEG | 1-D | ERS | Statistical features from the time and frequency domains | SVM, KNN, LDA | Valence – 62.37
Arousal – 62.37 |
| 24 | DREAMER | ECG | 1-D | ERS | Statistical features from the time, frequency, time-frequency domains, and nonlinear analysis-related | SVM | Valence – 86.09
Arousal – 87.80 |
| 25 | DREAMER | ECG | 1-D | ERS | Deep-learning | Convolutional Neural Network (CNN) | Valence – 74.90
Arousal – 77.10 |
| 12 | DREAMER | ECG | 1-D | ERS | Statistical features from the time and frequency domains | SVM | Valence – 65.80
Arousal – 65.40 |
| 26 | (MWM-HIT)
100 subjects |
ECG | 2-D | Authentication System | PQRST peaks | CNN | 99.99* |
| 27 | PhysioNet dataset (Fantasia and ECG-ID) | ECG | 2-D spectral | Authentication system | Spectrogram | CNN | 99.42* |
| 28 | Own dataset generated by FLUKE “ProSim 4 Vital Sign
and ECG Simulator” |
ECG | 2-D spectral | ECG classification | Instantaneous frequency and spectral entropy | LSTM | 100* |
| 4 | Zhejiang dataset | ECG | 2-D | Myocardial infarction screening | Object detection | DenseNet, KNN, SVM | DenseNet – 94.73*
KNN – 89.84* SVM – 92.19* |
| 29 | MIT-BIH arrhythmia dataset | ECG | 2-D spectral | Arrhythmia classification | Local features from 2-D images using deep learning | CNN | 99.11* |
| 5 | Physiobank dataset | ECG | 1-D and 2-D | Ventricular Arrhythmia detection | ECG beat images | SVM, Probabilistic Neural Network (PNN), KNN, Random Forest (RF) | 99.99* (both are useful) |
| 15 | Own dataset
11 subjects, 6 stimuli |
ECG and EEG | 2-D spectral | ERS | Statistical features from the time and frequency domains (R-R interval spectrogram) | CNN | ECG – 91.67
EEG – 90.00 |
| 30 | AMIGOS, DEAP | ECG, PPG, EDA | 2-D spectral | ERS | Features extracted from spectrogram by ResNet-50 | Logistic Regression | AMIGOS – 78.30
DEAP – 69.45 |