Multi-distance fluctuation based dispersion fractal for epileptic seizure detection in EEG signal

The developmental methods for evaluating the complexity of univariate signals has attracted extensive attention. Therefore, entropy was discovered to be one of the best methods for evaluating the complexity of a biological signal. Recent studies on signal complexity using fractal dimension have been able to tackle the domination of entropy measurement. It was found that Fluctuation-based dispersion entropy (FDispEn) is one of the recently proposed methods based on permutation (PE) and Shannon entropies (SE). This method analyzes the signal's uncertainty and deals with its fluctuations. FDispEn mainly calculates the differences between adjacent elements of the dispersion patterns based on Shannon entropy, however, it is limited by distance. Therefore, this study proposes a new feature extraction method based on FDispEn by expanding adjacent elements’ measurement distance using the multi-distance signal level differences (MSLD) method. MSLD is an upgrade of the gray-level difference (GLD) that is used to evaluate one-dimensional signals. Furthermore, it is used to calculate several distances of adjacent dispersion patterns. The MSLD is also applied in FDispEn to form multi-distance FDispEn (MFDispEn). Other signal complexity evaluations involving two fractal dimension methods, namely Higuchi and Katz's were used in forming the multi-distance fluctuation-based dispersion fractal (MFDF). The performance of FDispEn, MFDispEn, and three variations of MFDF were compared to evaluate the epileptic EEG signals. The results showed that the multi-distance application on MFDispEn and MFDF produced a better separability than FDispEn. Meanwhile, the MFDF outperformed the FDispEn and MFDispEn as it showed a higher accuracy, sensitivity, and specificity in classifying epileptic EEG signals. © 2021 Elsevier Ltd