Epileptic seizure prediction on EEG signal using multiscale intrinsic mode fluctuation-based dispersion entropy

Epileptic seizure prediction on EEG signal using multiscale intrinsic mode fluctuation-based dispersion entropy

Fluctuation-based dispersion entropy (FDE) has emerged as a robust method for analysing biological signals. To improve its adaptability across multiple time scales, this study introduces a novel feature extraction technique, Intrinsic Mode Fluctuation-Based Dispersion Entropy (IMFDE), which combines empirical mode decomposition (EMD) and multiscale FDE. The electroencephalogram (EEG) signals were pre-processed, decomposed into intrinsic mode functions (IMFs), which were then used to compute IMFDE features. The proposed method was validated on a short-term dataset (Bonn University) and long-term dataset (Temple University Seizure Corpus–TUSZ). On the Bonn dataset, IMFDE achieved 99% accuracy, 100% sensitivity, and 98% specificity, and thus selected for subsequent testing for seizure prediction on TUSZ. On the TUSZ, IMFDE maintained a low false prediction rate (0.53 h-1) and 100% sensitivity for 2- and 4-min seizure prediction horizons. These results demonstrate that IMFDE outperforms existing entropy-based methods, offers an adaptive and reliable approach for epileptic seizure forecasting.