Table 1 Summary of various seizure detection algorithms
From: Various epileptic seizure detection techniques using biomedical signals: a review
Authors | Year | Features/transforms used/channels | Classifiers | Window size | Dataset | Performance metrics (%) |
|---|---|---|---|---|---|---|
Runarsson et al. (T) | 2005 | Amplitude and time difference/single channel | SVM | Frames are not fixed | Self-recorded | Sensitivity: 90% |
Alejandro et al. (T) | 2017 | Energy/multichannel | Genetic algorithm/threshold | NA | CHB-MIT | False rate 0.39 per 24 h in average |
Turky et al. (T) | 2016 | Histograms | Threshold value | 10 s | CHB-MIT | 97.14 sens., 98.58 spec. |
Mursalin et al. (T) | 2017 | Mean, mode, etc./wavelets | Random forest (RF) | NA | Bonn dataset | Avg acc. 98.45 |
Michael | 2016 | FFT coff. | Random forest (RF) | 1 s | Self-rec. by compi. | NA |
Supratak et al. | 2014 | NA | NA | NA | CHB-MIT | Sensitivity: 100%, false positive rate: 7.98/h, detection latency: 6.87% |
Yoo et al. (T) | 2013 | Energy/multichannel | SVM | 2 s | CHB-MIT | Accuracy: 84.4%, classification |
Ahamad et al. | 2014 | NA | NA | NA | CHB-MIT | Sensitivity: 98.5%, detection latency: 1.76 s |
Kiranyaz et al. | 2014 | NA | NA | NA | CHB-MIT | Sensitivity: 89% |
Dalton et al. (T) | 2012 | Signature of seizure/single channel | Template matching | 12–25 s | Dataset of 21 seizure | Sensitivity: 91%, specificity: 84%, battery lifetime: 10.5 h |
Rana et al. (F) | 2012 | Phase slope Index/Fourier/multichannel | None | 10–60 s | EEG and ECoG time series, BW: 0.5–100 Hz | Accuracy: 100% |
Khamis et al. (F) | 2013 | Frequency moment signatures/Fourier/single channel | Powell’s direction set method | 32 s | 12 patients with six data records (R1 to R6) | Sensitivity: 91%, false positives per hour: 0.02 |
Acharya et al. (F) | 2012 | Four entropies/single channel | SVM, FSC, PNN, KNN, NBC, DT, and GMM | 23.6 s | Channel Self-recorded data | Average accuracy: 98.1% |
Zhou et al. (W) | 2012 | Lacunarity and fluctuation index on wavelet scales/multichannel Wavelet | Bayesian linear discriminant analysis (BLDA) classifier | 4 s without overlapping | Freiburg database of 21 patients | Sensitivity: 96.25%, false detection rate: 0.13/h, mean delay time: 13.8 s |
Hasan. et al. (w) | 2016 | Mean, maximum, minimum, standard deviation, and average power of absolute values of wavelet and Hilbert transform coefficients | K-nearest neighbourhood (kNN) | Bonn database | Accuracy: the wavelet 100 and 96% for the A–E and B–E datasets, respectively, the Hilbert: transform 100 and 100% for the A–E and B–E datasets, respectively | |
Liu et al. (W) | 2012 | Relative energy, relative amplitude, coefficient of variation, fluctuation index/multichannel/wavelet | SVM | 4 s | Dataset of 509 h from 21 epileptic patients | Sensitivity: 94.46%, specificity: 95.26%, false detection rate: 0.58/h |
Panda et al. (W) | 2010 | Energy, entropy, standard deviation/single channel/wavelet | SVM | 0.5 s | 500 epochs of EEG data from five different brain activities (100 signals per epoch) | Accuracy: 91.2% |
Guangyi et al. | 2017 | Amplitude of coefficients/wavelet/Fourier | KNN | Bonn database | Classification rates (100%) | |
Khan et al. (W) | 2012 | Energy and normalized coefficient of variation/single/wavelet | Simple LDA classifier | 25 s | Self-recorded | Sensitivity: 83%, specificity: 100%, accuracy: 92%, overall precision: 87% |
Rezvan et al. | 2017 | Maximum, minimum, average and standard deviation/single/wavelet | MLP Neural network | 26.3 s | Bonn database | confusion matrix: accuracy, 98.33 |
Wang et al. | Approximate entropy/single/wavelet | Neyman–Pearson criteria and SVM | 70–200 ms | Changhai Hospital database (scalp EEG) | Detection accuracy: 98%, false detection rate: 6% | |
Zainuddin et al. | 2012 | Maximum, minimum, standard deviation of absolute wavelet coefficient/single/wavelet | WNN | 23.6 s | Bonn database | Sensitivity: 98%, accuracy: 98% |
Niknazar et al. | 2013 | Time delay, embedding dimension/single/wavelet | ECOG | 23.6 s | Bonn database | Accuracy: 98.67% |
Daou and Labeau | 2014 | SPIHT codes/single/wavelet | No classifier | 1 s | Accuracy: 90% | |
Shoaib et al. | 2014 | Wavelet energy/multichannel/wavelet | SVM | 2 s | MIT database (scalp EEG | Sensitivity: 91–96%, latency: 4.7–5.3 s, false alarm rate: 0.17–0.3/h |
Zandi et al. | 2010 | Combined seizure index/single/wavelet | No classifier | 10–40 s | EEG recordings from 14 patients approximately 75.8 h with 63 seizures | Sensitivity: 90.5%, false detection rate: 0.51 h-1, median detection delay: 7 s |
Tafreshi et al. (E) | 2008 | Mean of the absolute of each IMF, wavelet feature/single/EMD | Neural Network | 4–6 s | Five patients of Freiburg database (scalp and iEEG) | Accuracy: 95% |
Orosco et al. | 2009 | Energies of the IMFs/single/EMD | No classifier | I h | 90 EEG segments acquired for nine patients | Sensitivity: 56%, specificity: 75%, positive predictive value: 61%, negative predictive value: 72% |
Guarnizo and Delgado | 2010 | Instantaneous frequency, amplitude of each EMD component, skewness, kurtosis, Shannon’s entropy/single/EMD | Linear Bayes classifier | Whole signal | Five groups with 100 single-channel registers sampled at 173.61 Hz | Accuracy: 98% |
Alam and Bhuiyan | 2011,2013 | Skewness, kurtosis, variance, largest Lyapunov exponent, correlation dimension, approximate entropy from intrinsic mode/single/EMD | ANN | 23.6 s | Bonn database | Accuracy: 100% |
Bajaj and Pachori | 2014 | Modified central tendency measure/single/EMD | No classifier | 15 s | Freiburg database for 21 patients (scalp and iEEG) | Sensitivity: 90%, specificity: 89.31%, error detection rate: 24.25% |
Sabrina et al. | 2016 | Euclidian, Bhattacharya and kolomogorov (IMFs) | PHA(unsupervised) | 2 s | CHB-MIT | Accuracy of 98.84% |
Dattaprasad et al. | 2017 | Coefficients of Hilbert(IMF)/EMD/Hilbert | ANN | 23.6 s | Bonn database | Accuracy of 96% |
Kaveh et al. | 2014 | Absolute mean, median, standard deviation, maximum and minimum value of the coefficients/rational STFT | MLP | 1.5 s | Bonn database | Accuracy: E–A: 99.8 E–B: 99.3, etc. |
Kaveth and Peter | LGBP/1 D LBP/multiple channel | SVM/RF/Log-reg | 1 | CHB-MIT | Sensitivity of 70.4% and the overall specificity of 99.1% |