Late Fusion of Time-Domain and Time-Frequency Sleep EEG for Future Cognitive Impairment Prediction

Nahyun Lee1, Kim TaeHwan2, TaeYoung Kang2, DongJae Lee3, DONGGUN KIM4, Jeongbin You5, younghyuk kim4, Seungsang Oh4, Wuming Gong6, Il-Youp Kwak2
1Chung Ang University, 2Chung-Ang University, 3¹Department of Anesthesiology and Pain Medicine, Korea University Guro Hospital, Seoul, Republic of Korea ²Department of Physiology, Korea University College of Medicine, Seoul, Republic of Korea, 4korea university, 5Koera University, 6University of Minnesota

Abstract

For the ``Screening for Cognitive Impairment During Sleep Studies: The George B. Moody PhysioNet Challenge 2026,'' Team CAU_KU develops a compact late-fusion system to predict cognitive impairment one to six years after clinical polysomnography (PSG). Heterogeneous montages and acquisition conditions complicate cross-site prediction, so we focus on electroencephalography (EEG), which can be consistently harmonized across sites and captures neural activity relevant to cognitive decline. Our pipeline standardizes six EEG derivations from variable PSG configurations and samples quality-controlled nighttime windows. Time-domain and time-frequency branches are trained separately, calibrated at the participant level, and fused by probability averaging. Both branches are trained from scratch, yielding a compact solution without pretrained models. These results suggest that complementary time-domain and time-frequency sleep EEG representations support cross-site prediction of future cognitive impairment.