Automatic Seizure Detection Using Multi-Input Deep Feature Learning Networks for EEG Signals

Table 5

Comparisons with state-of-the-art seizure detection methods using the same dataset.

Authors	Methodology	Dataset	Sensitivity(%)–specificity(%)–accuracy(%)
Truong et al. [21]	STFT, CNN	CHB-MIT	81.2–NR–NR
Tian et al. [22]	Deep multi-view feature learning, FFT, WPD, CNN	CHB-MIT	96.7–99.1−98.3
Li et al. [39]	Unified temporal-spectral squeeze-and-excitation network	CHB-MIT	92.41–96.05−95.96
	—	Bonn	NR–NR–99.8
Peng et al. [40]	Stein kernel-based sparse representation	CHB-MIT	97.85–98.57−98.21
	—	Bonn	98.43–98.67−98.67
Deng et al. [41]	Transductive transfer learning fuzzy system	CHB-MIT	97.16–97.03−97.15
Hossain et al. [42]	Raw EEG as 2D Array, CNN	CHB-MIT	90–91.65−98.05
Shoeibi et al. [43]	Handcrafted features, SVM, KNN, CNN	Bonn	NR–NR–99.53
Li and Chen [44]	FFT, 2D matrix, SVM	CHB-MIT	98.28–98.5−98.47
Jiang et al. [45]	Personal correlation coefficient and mutual information, SVM	CHB-MIT	97.72–95.62−96.67
Zhao et al. [46]	CNN, transformer	CHB-MIT	97.70–97.6−98.76
Thara et al. [47]	DNN, Different feature scaling techniques	Bonn	98.59–91.47−97.21
Xiong et al. [48]	Multivariate variational mode decomposition, RF	CHB-MIT	98.24–97.83−97.39
Ein Shoka et al. [25]	Transfer learning, CNN	CHB-MIT	88.89–84.21−86.11
Akyol [49]	Stacking ensemble approach, deep neural networks (DNN)	Bonn	93.11–98.18−97.17
Pan et al. [50]	Hybrid input, DWT, STFT, FFT, CNN	Bonn	NR–NR–97.89
This work	Multi-view features, MDFLN	CHB-MIT	98.42–97.77−98.09
	—	Bonn	97.02–98.71−98.4

The NR stands for not reported values.