Harnessing Long-Range Temporal Correlations for Advanced Epilepsy Classification

Abstract

Listen

Translate article

In this study, we suggest that detrended fluctuation analysis may reveal alternations in long-range temporal correlations associated with epileptic status. We consider two groups of subjects: patients with confirmed focal epilepsy, and healthy controls. Both groups were exposed to intermittent photic stimulation. Analysis based on event-related spectral perturbations revealed that photic driving in epileptic patients is higher in the α band and lower in other bands. This result is related to altered excitability in nonphotosensitive epilepsy. To prove this, we tested two hypotheses. First, we assess long-range temporal correlations using detrended fluctuation analysis, with the objective of evaluating whether this metric differs between epileptic patients and healthy controls through the application of between-subject statistics. Second, we investigate whether detrended fluctuation analysis provides more valuable insights into the aforementioned differences in comparison to traditional spectral analysis of brain signals. More precisely, we test if a machine learning algorithm trained on detrended fluctuation analysis-based features outperforms one trained on the spectral-based features in classifying between epileptic patients and controls. Furthermore, we study whether the features differ between classifiers by implementing feature importance assessment. Our findings demonstrate that the classifier based on detrended fluctuation analysis exhibits higher efficiency, and its features are notably distinct from those of the classifier based on spectral analysis. We postulate that the long-range temporal correlations captured via detrended fluctuation analysis reveal novel aspects of the epileptic brain response to intermittent photic stimulation, and they are more pronounced than the features captured via spectral analysis. Published by the American Physical Society 2025