Estimating cognitive function score from mild cognitive impairment to moderate dementia using a hybrid model combining plasma biomarkers with electroencephalogram signal

BackgroundPredicting cognitive function across dementia stages remains challenging. Plasma biomarkers and electroencephalogram (EEG) features may provide complementary information, but their combined predictive value requires further study.ObjectiveTo evaluate the feasibility of integrating plasma biomarkers and EEG features to predict cognitive function in dementia and examine their correlations.MethodsFrom September 2023 to October 2024, 75 patients from two medical centers with mild cognitive impairment, mild dementia, or moderate dementia were enrolled. Resting-state 19-channel EEG data yielded 2737 time-frequency and connectivity features. Plasma biomarkers included tau, p-Tau181, Aβ_1-42, neurofilament light chain (NfL), brain-derived neurotrophic factor, apolipoprotein E genotype, and glial fibrillary acidic protein. Cognitive function was assessed using Cognitive Abilities Screening Instrument (CASI), Mini-Mental State Examination, Montreal Cognitive Assessment (MoCA), and Clinical Dementia Rating Sum of Boxes. Machine learning models were developed using plasma-only, EEG-only, and hybrid approaches.ResultsNfL was negatively correlated with CASI (t = -2.059, p < 0.05). Several EEG features showed moderate correlations with cognitive measures and plasma biomarkers, with delta-band relative power between left frontal and temporal regions (F7-FT7) showing the strongest correlation with MoCA. The hybrid model achieved the best performance, with R² > 0.74 across all cognitive measures, outperforming plasma-only and EEG-only models.ConclusionsIntegrating EEG features with plasma biomarkers improves prediction of cognitive function from mild cognitive impairment to moderate dementia, pending external validation.