Comprehensive Summary
This study considers how EEG microstates and power spectral characteristics differ in people with epilepsy (PWE) with mild cognitive impairment (MCI) compared to people without MCI. EEG data from subjects were analyzed retrospectively in 627 people, and the microstate dynamics and spectral power were examined in relation to five frequency bands. Machine learning techniques of Support Vector Machine, Neural Network, Random Forest, K-Nearest Neighbors, and Naïve Bayes were used to determine predictive models for MCI comorbidity. The best model was found using a neural network with the microstate variables, area under the ROC curve (AUC) found to be 0.93, an accuracy 89% excellent calibration. The EEG analysis showed that the PWE with MCI showed a significant difference in microstates A-D, particularly in relation to duration, occurrence, and transition probabilities, along with significant differences in delta band spectral power. Points to note in the discussion include how EEG markers could be considered indicative of impaired functional connectivity between auditory, visual, and executive brain networks that may be involved in cognitive decline in epilepsy, and which of these could be used as electrophysiological markers to show cognitive decline in the MCI processes of PWE.
Outcomes and Implications
This study is clinically relevant in that it provides a non-invasive and objective tool for early detection of cognitive impairment in epilepsy in patients who cannot complete cognitive screen tasks of the traditional type e.g., MoCA or MMSE. The predictions of the neural network model are more accurate and have greater clinical applicability than younger people's standard screening processes, thereby increasing the possibility for intervention and tracking of cognitive decline. The authors corroborated that a larger multicenter study and validation were needed for the neural network model, but EEG predictive modelling could be included in the clinical processes that the operational workflow illustrates, with possible future personalisation of care and addressing cognitive decline in the PWE population.