EEG network features and APOE ε4 for amyloid classification in mild cognitive impairment

Resting-state electroencephalography (EEG) may provide information complementary to apolipoprotein E ε4 (APOE ε4) for identifying amyloid pathology in mild cognitive impairment. We evaluated this hypothesis in an exploratory cross-sectional study of 93 patients, including 54 with amyloid-positive positron emission tomography, using EEG functional connectivity, graph features and 11 classifiers under nested cross-validation. In the original partition, extreme gradient boosting with all domains achieved an area under the receiver operating characteristic curve (AUC) of 0.806 versus 0.682 for its APOE-only model (paired difference 0.124, 95% confidence interval 0.010–0.238).

The median paired difference across 30 additional partitions was 0.088. Among non-carriers, the combined model identified 11 of 16 amyloid-positive patients, with 10 false positives among 31 amyloid-negative patients. The gain was smaller against carrier status used directly (difference 0.057, −0.043–0.157).

Nesting classifier selection reduced the median AUC to 0.734, and alternative EEG epoch samples yielded AUCs of 0.571–0.773 on the original partition. These exploratory findings support further evaluation of complementary EEG and genetic information, while demonstrating dependence on the comparator, model selection and epoch sampling. Reliable EEG estimation and external validation are required before clinical use.

Area under the receiver operating characteristic curve Seoul Neuropsychological Screening Battery The authors thank the patients and their caregivers for their participation. No artificial intelligence tool was used in a way that would meet the journal’s authorship criteria; any use was limited to language polishing of author-drafted text.This work was supported by the Bio & Medical Technology Development Program of the National Research Foundation of Korea (NRF), funded by the Korea government (Ministry of Science and ICT, MSIT) [grant number RS-2025-02217919], and by Korea University Anam Hospital and Korea University College of Medicine [grant number K2606611]. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.

This work was supported by the Bio & Medical Technology Development Program of the National Research Foundation of Korea (NRF), funded by the Korea government (Ministry of Science and ICT, MSIT) [grant number RS-2025-02217919], and by Korea University Anam Hospital and Korea University College of Medicine [grant number K2606611]. Department of Neurology, Korea University Anam Hospital, Korea University College of Medicine, Seoul, 02841, Republic of Korea Jooheon Kong, Mingyeong So, Inyoung Lee, Hyunsung Park, Hayom Kim, Chan-Nyoung Lee & Jung Bin Kim Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Boston, USA Department of Neurology, Korea University Guro Hospital, Korea University College of Medicine, Seoul, Republic of Korea KU-KIST Graduate School of Converging Science and Technology, Korea University, Seoul, Republic of Korea Correspondence to Chan-Nyoung Lee or Jung Bin Kim. The authors declare no competing interests.

Large language model-based tools assisted with editing the revised manuscript, preparing analysis code and checking reporting consistency. The authors are responsible for reviewing the text, code and scientific interpretation and for the final submitted content. Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.