TY - GEN

T1 - Multi-Modal Fusion of EEG and Genetic Markers for Depression Prediction from ICBrainDB

AU - Firoz, Neda

AU - Berestneva, Olga Grigorievna

AU - Savostyanov, Alexander

AU - Aksyonov, Sergey Vladimirovich

N1 - Conference code: 17

PY - 2025/12/18

Y1 - 2025/12/18

N2 - Major depressive disorder (MDD) remains a leading cause of disability worldwide, yet current diagnostic approaches rely heavily on subjective clinical assessments, which can delay intervention and reduce diagnostic reliability. Objective, multimodal biomarkers offer a promising route toward earlier and more accurate detection. Electroencephalography (EEG) captures real-time neural dynamics, while genetic biomarkers provide stable indicators of molecular predisposition, offering complementary perspectives on depression risk. This study introduces EEG–Gene Fusion Depression Network (EGF-DepNet), the first end-to-end deep learning framework to integrate EEG-derived features and gene-based biomarkers for depression prediction. Two fusion strategies are proposed: (i) an Attention Fusion model employing cross-modal MultiHeadAttention within a compact Conv1D network, and (ii) a Transformer Fusion model that encodes EEG and genomic embeddings as tokens in a lightweight self-attention encoder. Using the ICBrainDB dataset, both architectures achieved high predictive performance, with Transformer Fusion outperforming Attention Fusion across multiple evaluation metrics, including F1-score (0.727 vs. 0.600) and AUC (0.845 vs. 0.749). Results demonstrate that multimodal EEG–genomic integration improves classification robustness over unimodal approaches, effectively leveraging the temporal sensitivity of EEG and the trait stability of genetic markers. This work advances the development of biologically informed, AI-driven diagnostic tools, offering a pathway toward more precise, scalable, and personalized approaches in precision psychiatry.

AB - Major depressive disorder (MDD) remains a leading cause of disability worldwide, yet current diagnostic approaches rely heavily on subjective clinical assessments, which can delay intervention and reduce diagnostic reliability. Objective, multimodal biomarkers offer a promising route toward earlier and more accurate detection. Electroencephalography (EEG) captures real-time neural dynamics, while genetic biomarkers provide stable indicators of molecular predisposition, offering complementary perspectives on depression risk. This study introduces EEG–Gene Fusion Depression Network (EGF-DepNet), the first end-to-end deep learning framework to integrate EEG-derived features and gene-based biomarkers for depression prediction. Two fusion strategies are proposed: (i) an Attention Fusion model employing cross-modal MultiHeadAttention within a compact Conv1D network, and (ii) a Transformer Fusion model that encodes EEG and genomic embeddings as tokens in a lightweight self-attention encoder. Using the ICBrainDB dataset, both architectures achieved high predictive performance, with Transformer Fusion outperforming Attention Fusion across multiple evaluation metrics, including F1-score (0.727 vs. 0.600) and AUC (0.845 vs. 0.749). Results demonstrate that multimodal EEG–genomic integration improves classification robustness over unimodal approaches, effectively leveraging the temporal sensitivity of EEG and the trait stability of genetic markers. This work advances the development of biologically informed, AI-driven diagnostic tools, offering a pathway toward more precise, scalable, and personalized approaches in precision psychiatry.

KW - электроэнцефалография

KW - генетические биомаркеры

KW - мультимодальная интеграция

KW - прогнозирование депрессии

KW - трансформер

KW - модели внимания

KW - EEG

KW - Genetic biomarker

KW - Multimodal fusion

KW - Depression prediction

KW - Transformer

KW - Attention Models

UR - https://www.scopus.com/pages/publications/105031777048

UR - https://www.mendeley.com/catalogue/de5d9a7a-8c7e-3a6b-b659-468f62297df5/

U2 - 10.1109/apeie66761.2025.11289224

DO - 10.1109/apeie66761.2025.11289224

M3 - Conference contribution

SN - 979-8-3315-5917-5

SP - 1

EP - 9

BT - 2025 IEEE XVII International Scientific and Technical Conference on Actual Problems of Electronic Instrument Engineering (APEIE)

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2025 IEEE XVII International Scientific and Technical Conference on Actual Problems of Electronic Instrument Engineering

Y2 - 14 November 2025 through 16 November 2025

ER -