Today, electroencephalography is used to measure brain activity by creating signals that are viewed on a monitor. These signals are frequently used to obtain information about brain neurons and may detect disorders that affect the brain, such as epilepsy. Electroencephalogram (EEG) signals are however prone to artefacts. These artefacts must be removed to obtain accurate and meaningful signals. Currently, computer-aided systems have been used for this purpose. These systems provide high computing power, problem-specific development, and other advantages. In this study, a new clinical decision support system was developed for individuals to detect epileptic seizures using EEG signals. Comprehensive classification results were obtained for the extracted filtered features from the time-frequency domain. The classification accuracies of the time-frequency features obtained from discrete continuous transform (DCT), fractional Fourier transform (FrFT), and Hilbert transform (HT) are compared. Artificial neural networks (ANN) were applied, and back propagation (BP) was used as a learning method. Many studies in the literature describe a single BP algorithm. In contrast, we looked at several BP algorithms including gradient descent with momentum (GDM), scaled conjugate gradient (SCG), and gradient descent with adaptive learning rate (GDA). The most successful algorithm was tested using simulations made on three separate datasets (DCT_EEG, FrFT_EEG, and HT_EEG) that make up the input data. The HT algorithm was the most successful EEG feature extractor in terms of classification accuracy rates in each EEG dataset and had the highest referred accuracy rates of the algorithms. As a result, HT_EEG gives the highest accuracy for all algorithms, and the highest accuracy of 87.38% was produced by the SCG algorithm.

Epilepsy is one of the most common neurological disorders in the world [

The use of databases is an important approach to the analysis of EEGs. In the literature, different computer-aided models using different EEG signal databases have been proposed. A summary of the studies using the BONN EEG database in the literature is presented in

Authors | Models | Dataset and accuracy rate (%) | |||||
---|---|---|---|---|---|---|---|

[ |
FSC, SVM, KNN, PNN, DT, GMM, NBC | Z-N-S: 98.1 | |||||

[ |
EMD, HOM, ANN | AB-CD-E: 80 | A-D-E: 100 | A-E: 100 | D-E: 100 | ABCD-E: 100 | |

[ |
AB-E: 100 | C-E: 100 | CD-E: 100 | ABCD-E: 100 | |||

[ |
DTCWT, Fourier features, k-NN | A-E: 100 | AB-E: 100 | CD-E: 100 | ABCD-E: 100 | ||

[ |
A-E: 100 | ABCD-E:100 | AB-CD-E: 96.28 | A-D-E:100 | D-E:100 | C-E:100 | |

[ |
GA, SVM | A-E:100 | |||||

[ |
Star graph topological indices, GDA | Z-S: 99.7 | ZONF-S: 98.6 | ||||

[ |
DWT, RWE, ANN | Z-S: 95.2 | |||||

[ |
DWT, MLPNN | Z-S: 100 | ZNF-S: 97.75 | FNOZ-S: 97.77 | |||

[ |
Multiwavelet trans. approximate entropy—MLPNN | ZONF-S: 98.27 | |||||

[ |
C-E: 95.33 | ||||||

[ |
FFT-QDA, AR-MLPNN | A-E: 99.78 | B-E: 99.55 | C-E: 99.62 | D-E: 99.46 | ||

[ |
CD, LLE, H, entropy, Surrogate data analysis | Z-S: 90 | |||||

[ |
Shannon’s entropy, ANFIS | Z-S: 92.22 | |||||

[ |
BayesNet, SVM, ANN, LR, FT | A-E: 99.5 | A-D: 99.5 | D-E: 95.5 | CD-E: 97 | AB-CDE: 93 | A-D-E: 95.67 |

[ |
DWT, ApEn, ANN, SVM | A-E: 100 | B–E: 92.5 | C-E: 100 | D-E: 95 | BCD-E: 94 | ABCD-E: 94 |

[ |
fApEn, |
A-E: 100 | B-E: 100 | C-E: 99.6 | D-E: 95.85 | ACD-E: 98.1 | BCD-E: 98.2 |

ABCD-E: 97.38 | |||||||

[ |
WT, phase space reconstruction, NEWFM | A-E: 98.17 | |||||

[ |
MODWT-based LND |
A-E: 100 | AB-CDE: 98.48 | ABCD-E: 99 | AB-CD-E: 98.1 | ||

[ |
TF, ApEn, linear or nonlinear classifiers, RBFSVM | ABCD-E: 98 | A-D-E: 98.67 | A-B-C-D-E: 85.9 | |||

[ |
A-E: 99.85 | ||||||

[ |
Improved correlation feature selection, RFC | A-E: 100 | B-E: 98 | C-E: 99 | D-E: 98.5 | ||

ACD-E: 98.5 | BCD-E: 97.5 | CD-E: 98.67 | ABCD-E: 97 | ||||

[ |
WT-based features, entropy, ANN, SVM | A-E: 99 | |||||

Approximate entropy, ANN, SVM | A-D-E: 96 | ABCD-E: 99 | AB-CD-E: 95 | A-B-C-D-E: 94 | |||

[ |
Permutation entropy, SVM | Z-S: 93.5 | O-S: 82.8 | N-S: 88 | F-S: 79.94 | FNOZ-S: 86.1 | |

[ |
DWT, ANN | A-E: 100 | ABCD-E: 99 | AB-CDE: 98 | A-D-E: 96.67 | AB-CD-E: 95.6 | |

[ |
A-E: 100 | ABCD-E: 99 | A-D-E: 99.3 | ACD-E: 99.28 | AB-CD-E: 98.37 | ||

[ |
FFT, DT | Z-S: 98.72 | |||||

[ |
DSTFT, BayesNet, LR, SVM, KNN | E-A: 99.8 | E-B: 99.3 | E-C: 98.5 | E-D: 94.9 | E-ABCD: 98 | |

[ |
ATFFWT, |
A-E: 100 | B-E: 100 | C-E: 99 | D-E: 98.5 | ||

AB-E: 100 | CD-E: 98.6 | AB-CD: 92.5 | ABCD-E: 99 | ||||

[ |
CT, LS-SVM | A-E: 99.9 | B-E: 96.3 | C-E: 96.2 | D-E: 93.6 | A-D: 84.9 | |

[ |
TF, RNN | Z-S: 99.6 | |||||

[ |
Z-S: 99.5 | N-F-S: 95 | Z-F-S: 97.5 | Z-O-F-N-S: 93 | |||

[ |
LBP, SVM | ZO-S: 100 | NF-S: 99.45 | ZO-NF-S: 98 | ZONF-S: 99 | ||

[ |
RFC, total and fractional energy, entropy | Z-O-N-F-S: 91 | ZO-NF-S: 98 | ||||

[ |
TF, ANN | Z-S: 100 | ZONF-S: 97.7 | Z-F-S: 99.28 | ZO-NF-S: 97 | ||

[ |
ANN, TF | Z-S: 100 | Z-F-S: 100 | Z-O-N-F-S: 89 | |||

[ |
P-1D, CNN | AB-CD-E: 99 | AB-CD: 99.9 | AB-E: 99.8 | A-E: 100 | B-E: 99.8 | CD-E: 99.7 |

D-E: 99.4 | BCD-E: 99.3 | BC-E: 99.5 | BD-E: 99.6 | AC-E: 99.7 | C-E: 99.1 | ||

ABCD-E: 99.7 | AB-CDE: 99.5 | ABC-E: 99.97 | ACD-E: 99.8 | ||||

[ |
Symlets wavelets, statistical mean energy std, PCA, GBM-GSO, RF, SVM | Z-S: 100 | O-S: 100 | N-S: 98.4 | F-S: 98.1 | OZ-S: 100 | NF-S: 98.1 |

OZ-NF: 93.2 | FNOZ-S: 98.4 | FN-OZ-S: 96.5 | |||||

[ |
Wavelet-based sparse functional linear mode | A-E: 100 | ABCD-E: 100 | ||||

[ |
WPD-based |
A-E: 100 | B-E: 99.94 | C-E: 99.85 | D-E: 99.38 | AB-E: 99.97 | CD-E: 99.58 |

AB-CDE: 99 | ABCDE: 99.7 | A-D-E: 99.39 | AB-CD-E: 98 | ||||

[ |
FWHVA, k-NN | A-E: 100 | D-E: 93 | ABCD-E: 95.4 | |||

[ |
ABCDE: 100 |

The most commonly used models in the BONN database studies are artificial neural networks (ANN), support vector machine (SVM), k-nearest neighbors (k-NN), and recursive flow classification (RFC) (

There are some studies in which these methods achieve 100% success. However, the reported studies do not have the same datasets. The accuracy rate obtained for the method recommended for the problem of classifying Z, O and N, F, S signals, which is needed by clinical experts in our study, is 87.38%. It is the method with the second-best classification accuracy in the literature for this data set. The best result is 99.5% obtained by Ullah et al. [

In this study, three ANN back propagation (ANN-BP) algorithms were used to classify EEG signals to determine the best feature extractor and algorithm. The steps in our study can be summarized as follows. (i) Finite impulse response (FIR) filtering was used for the preprocessing to remove noise from the EEG signals;

(ii) The time-frequency domain features were extracted by discrete continuous transform (DCT), fractional Fourier transform (FrFT), and Hilbert transform (HT);

(iii) The features were obtained from the DCT_EEG, FrFT_EEG, and HT_EEG datasets;

(iv) DCT_EEG, FrFT_EEG, and HT_EEG were classified with the gradient descent with momentum (GDM), scaled conjugate gradient (SCG), and gradient descent with adaptive learning rate (GDA) training algorithms for the extracranial and the intracranial EEG signals; and

(v) Classification accuracy rates were compared for the training algorithms according to the best time-frequency features.

The rest of the paper is organized as follows. The methods of the proposed models are described step by step in Section 2. The experimental results are given in Section 3, and the conclusions of the study are presented in Section 4.

In this study, the extracranial and intracranial EEG signals were used for classifying the features of the significant time-frequency EEG signals from the ANN-BP algorithms.

The analyzed EEG signals were obtained from the publicly available BONN database [

Preprocessing is a crucial step for the removal of artifacts from EEG signals before extracting significant signal features. Therefore, in this study the FIR filtering method was used to remove artifacts.

FIR filtering has a non-recursive impulse response that has a finite duration of

In this study, the structure of the FIR filter for preprocessing shows that the impulse response of

So,

The FIR filtering structure is shown in

The Kaiser window is crucial for reducing spectral leakage in the analysis of EEG signals that concentrate most of the energy in the amplitude. It is almost optimal, and it depends on the parameter

_{0} shows the zero order Bessel function, which is measured using the power series expansion as in

In this study, significant features for extracranial and intracranial EEG signals were extracted by the time-frequency domain using DCT, FrFT, and HT. These extractions helped to describe significant features of EEG signal components that tend to be complex and chaotic structures. Three datasets were extracted by the time-frequency methods, and three different ANN-BP training algorithms were applied to compare classification accuracy rates.

DCT is given as an even function

The purpose of FrFT is transferring signals from the time domain to the frequency domain and to determine the most significant features for the EEG signals. The FrFT of EEG signals,

where

The variable parameter

In this study, it was extracted from filtered EEG signals time-domain HT relations where there were no poles on the

ANN-BP training algorithms are the most widely used algorithms for weight-updating strategies in classification processes [

The EEG datasets obtained from time-frequency methods (DCT, FrFT, and HT) were classified by ANN-BP algorithms.

Initially, the weights

Three basic training algorithms (GDM, SCG, GDA) were used to show the best classification performances with the effective time-frequency feature descriptor method.

The GDM algorithm allows the neural network model to respond to both local degradation and recent trends on the error surface. The momentum performs as a low-pass filter, which allows the minor features to be ignored on the error surface of the neural network. The learning rate

where

The SCG can train any network as long as its weight and net input. SCG is an effective and fully automated optimization approach for the supervised learning algorithm that represents performance benchmarked against that of the standard ANN-BP. It does not add any user-dependent parameters that are crucial for its success. The algorithm avoids time consuming line search as per the learning iteration and uses a step-size scaling mechanism. The training step size equals the minimum quadratic polynomial fitted to

Choose the weight vector

Set

If the success is equal to true, then calculate the second order information as

If

Calculate the step size as

Calculate the comparison parameter.

If

If the

If the steepest descent direction is not equal to 0, then set

The output and error rate of the GDA algorithm are calculated in the neural network model. In each epoch, the new

In each epoch, the learning rate increases by the

Our proposed models were evaluated by computing the statistical parameters of Cohen’s Kappa coefficient and receiver operating characteristic (ROC).

The Kappa Test is a statistical method that measures the reliability of compliance between two or more observers. If the test is between two observers, it is called

An earlier study [

<0: harmony depends only on chance;

0.01–0.20: insignificant compliance;

0.21–0.40: poor compliance;

0.41–0.60: moderate compliance;

0.61–0.80: good fit;

0.81–1.00: very good level of the fit.

An ROC curve is a graphical plot that shows the classification ability for binary classification. The ROC curve is constructed by plotting the false positive rate (FPR) versus the true positive rate (TPR) for the various threshold settings.

Condition positive | Condition negative |
---|---|

True positive (TP) | False positive (FP) |

False negative (FN) | True negative (TN) |

The ROC curve can be generated by plotting the cumulative distribution function of the detection probability in the

In this study, the experiments were performed by using three different EEG signals datasets obtained using the DCT, FrFT, and HT for extracting the significant time-frequency EEG signal features. The experimental research consisted of the following steps:

The flowchart for obtaining the DCT_EEG dataset is shown in

Algo. | DCT_EEG dataset | FrFT_EEG dataset | HT_EEG dataset | |||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|

Epoch | Performance | Gradient | Accuracy (%) | Epoch | Performance | Gradient | Accuracy (%) | Epoch | Performance | Gradient | Accuracy (%) | |

57 | 0.303 | 0.208 | 68.94 | 1000 | 0.211 | 0.347 | 73.18 | 310 | 0.146 | 0.095 | 83.62 | |

21 | 0.161 | 0.059 | 82.39 | 78 | 0.130 | 0.067 | 82.56 | 50 | 0.008 | 0.085 | 87.38 | |

77 | 0.153 | 0.269 | 83.44 | 19 | 0.094 | 0.135 | 82.88 | 87 | 0.105 | 0.294 | 83.57 |

In this study, DCT_EEG, FrFT_EEG, and HT_EEG were obtained according to the following steps.

The training, test, and validation performance results of the algorithms are shown in

Datasets | GDM | SCG | GDA |
---|---|---|---|

DCT_EEG | |||

TPR = 0.62; FPR = 0.27 | TPR = 0.76; FPR = 0.13 | TPR = 0.76; FPR = 0.11 | |

FrFT_EEG | |||

TPR = 0.68; FPR = 0.24 | TPR = 0.79; FPR = 0.15 | TPR = 0.79; FPR = 0.15 | |

HT_EEG | |||

TPR = 0.80; FPR = 0.14 | TPR = 0.80; FPR = 0.07 | TPR = 0.77; FPR = 0.11 | |

The results of the proposed method were compared with other methods in the literature. In our study, the experimental results were compared with their classification accuracy rates and statistical analysis results. Hence, the proposed methods listed in

The

The plots of the nine confusion matrices mentioned earlier in the ROC curves are shown in

In this study, a novel clinical decision support system was developed for the diagnosis of epilepsy using extracranial and intracranial EEG signals. The main contribution of this study is that it proposes a brand-new computer vision-based approach for the measurement of EEG signals in epileptic individuals. Significant features were extracted using the time-frequency methods of DCT, FrFT, and HT. The extracted features were fed into the GDM, SCG, and GDA training algorithms. HT gave the best classification accuracy rates compared with DCT and FrFT methods with values of 87.38%, 83.62%, and 83.57%, respectively, for the three algorithms. The most distinctive time-frequency features were obtained using the significant EEG signal properties obtained from HT when applied to the SCG training algorithm. In future work, various features can be used to extract more efficient epilepsy-related properties, and will be tested for effectiveness. In particular, it is planned to use fractal-related, wavelet-related, and entropy-related features. In addition, more EEG signals data will be used to re-validate the novel learning algorithms, and other advanced machine learning algorithms will be validated with the ANN-BP training algorithms.