상세 보기
EMMNet: Learning Complementary Temporal and Structural Representations from EEG and MRI for Early Neurological Disorder Diagnosis
- Lee, Injae;
- Park, Jin-Hwi;
- Jo, Hyeonseo;
- Yoon, Young Chul;
- Paik, Joonki
WEB OF SCIENCE
0SCOPUS
0초록
Early and reliable diagnosis of neurological disorders, such as Alzheimer’s disease (AD) and mild cognitive impairment (MCI), remains a critical clinical challenge due to the heterogeneous nature of neurodegenerative conditions and the reliance on multiple complementary diagnostic tests. Recent studies have explored AI-based diagnostic models using Electroencephalography (EEG) and Magnetic Resonance Imaging (MRI); however, effectively leveraging their complementary characteristics remains non-trivial. EEG provides non-invasive and temporally rich measurements of brain activity but is highly susceptible to noise and inter-subject variability. MRI, in contrast, offers high-resolution structural information but lacks temporal sensitivity to functional brain dynamics. To address these inherent limitations, we introduce EEG–MRI Multi-modal neural network (EMMNet), a novel end-to-end deep learning framework that jointly exploits temporal EEG signals and structural MRI volumes for early neurological disorder diagnosis. EMMNet employs a dedicated temporal encoder for EEG and a spatial encoder for 3D MRI, with late-stage cross-modal fusion designed to promote complementary and modality-specific feature learning while mitigating cross-modal interference. To support multimodal learning under clinically realistic settings, we further present the CAUEMM dataset, the largest clinically labeled dataset comprising synchronized EEG and MRI recordings acquired from the same subjects. Extensive experimental evaluations demonstrate that EMMNet consistently and significantly outperforms single-modality baselines, highlighting the synergistic benefit of integrating EEG-derived temporal dynamics with MRI-based structural patterns. Overall, this work establishes a scalable multimodal diagnostic pipeline and provides a practical foundation for non-invasive, data-driven assessment of neurodegenerative disorders in clinical environments.
키워드
- 제목
- EMMNet: Learning Complementary Temporal and Structural Representations from EEG and MRI for Early Neurological Disorder Diagnosis
- 저자
- Lee, Injae; Park, Jin-Hwi; Jo, Hyeonseo; Yoon, Young Chul; Paik, Joonki
- 발행일
- 2027
- 유형
- Proceedings Paper
- 권
- 16824
- 페이지
- 279 ~ 293
- 언어
- ENG
- 출판사
- Springer Science and Business Media Deutschland GmbH
- 발행국가
- 미국
- 분량
- 15 페이지
- ISSN
- E 1611-3349
P 0302-9743