Advanced search
Publication Cover
Neuropsychiatric Disease and Treatment Volume 14, 2018 - Issue
Submit an article Journal homepage
95
Views
9
CrossRef citations to date
0
Altmetric
Original Research

Automatic epilepsy detection using fractal dimensions segmentation and GP–SVM classification

Jakub Jirka1 Department of Cybernetics and Biomedical Engineering, Faculty of Electrical Engineering and Computer Science, VSB – Technical University of Ostrava, Ostrava Poruba, Czech Republic
,
Michal Prauzek1 Department of Cybernetics and Biomedical Engineering, Faculty of Electrical Engineering and Computer Science, VSB – Technical University of Ostrava, Ostrava Poruba, Czech Republic
,
Ondrej Krejcar2 Center for Basic and Applied Research, Faculty of Informatics and Management, University of Hradec Kralove, Hradec Kralove, Czech Republic, [email protected]
&
Kamil Kuca2 Center for Basic and Applied Research, Faculty of Informatics and Management, University of Hradec Kralove, Hradec Kralove, Czech Republic, [email protected];3 Biomedical Research Center, University Hospital Hradec Kralove, Hradec Kralove, Czech Republic, [email protected]Correspondence[email protected]
Pages 2439-2449 | Published online: 21 Sep 2018

Figures & data

Figure 1 Result of automatic segmentation using fractal dimension on occipital area electrode signal from a patient with Jeavons syndrome.

Notes: A myoclonic epileptic seizure is detected and marked in the incipient segment. The upper graph represents the original EEG signal, the middle graph represents FDFV, and the lower one represents adaptive segmentation function. X-axis represents time in seconds. “U (t)” represents O1 occipital electrode O1 voltage data. “G” represents adaptive segmentation function as defined by Vari for the occpital electrode O1 voltage data.
Abbreviations: EEG, electroencephalography; FDFV, fractal dimension feature vector.
Figure 1 Result of automatic segmentation using fractal dimension on occipital area electrode signal from a patient with Jeavons syndrome.

Figure 2 Schematic flow chart for set extraction and further processing (compression, classification using GP+SVM).

Abbreviations: DWT, discrete wavelet transformation; EEG, electroencephalography; GP, genetic programming; SVM, support vector machine.
Figure 2 Schematic flow chart for set extraction and further processing (compression, classification using GP+SVM).

Figure 3 Population tree of suggested algorithm with two F functions (nodes) that make up new flagged vectors that are expressed by the offspring.

Note: In the illustrated example, two new flag vectors are created, the first being expressed as the sum of x5×x9 and the second as x3+x8.
Figure 3 Population tree of suggested algorithm with two F functions (nodes) that make up new flagged vectors that are expressed by the offspring.

Figure 4 GP–SVM block scheme.

Abbreviations: DWT, discrete wavelet transformation; GP, genetic programming; SVM, support vector machine.
Figure 4 GP–SVM block scheme.

Table 1 Clinically diagnosed epilepsy types of patients

Table 2 Frequency bands for four-level DWT decomposition

Figure 5 Four-level DWT decomposition for interictal EEG.

Abbreviations: DWT, discrete wavelet transformation; EEG, electroencephalography.
Figure 5 Four-level DWT decomposition for interictal EEG.

Figure 6 Four-level DWT decomposition for ictal EEG.

Abbreviations: DWT, discrete wavelet transformation; EEG, electroencephalography.
Figure 6 Four-level DWT decomposition for ictal EEG.

Table 3 Basic set feature vector mapping

Figure 7 Training stage of population tree samples separation. Population tree descendants combining the input features x9+x10 are clearly distinguishable from the population member genetically combining the input features x17+x18.

Notes: These two members (found using fitness function value) are suitable for the new features. Both the functions represented at X and Y axes are dimensionless.
Figure 7 Training stage of population tree samples separation. Population tree descendants combining the input features x9+x10 are clearly distinguishable from the population member genetically combining the input features x17+x18.

Figure 8 Confusion matrix for dataset shown in .

Notes: Diagonal elements represent the number of points correctly classified (predicted label = target true label) and off-diagonal elements express mislabeled samples. The higher the diagonal values mean, the better is the result of the classifier.
Figure 8 Confusion matrix for dataset shown in Figure 7.

Table 4 Classification accuracy

Table 5 Compressed feature vector with compression level 5

Table 6 Compressed and basic feature vectors-based classification performance comparison

Related research

People also read lists articles that other readers of this article have read.

Recommended articles lists articles that we recommend and is powered by our AI driven recommendation engine.

Cited by lists all citing articles based on Crossref citations.
Articles with the Crossref icon will open in a new tab.