NeuroBridge: Bio-Inspired Self-Supervised EEG-to-Image Decoding via Cognitive Priors and Bidirectional Semantic Alignment

Wenjiang Zhang; Sifeng Wang; Yuwei Su; Xinyu Li; Chen Zhang; Suyu Zhong

cs.CVNov 10, 2025

NeuroBridge: Bio-Inspired Self-Supervised EEG-to-Image Decoding via Cognitive Priors and Bidirectional Semantic Alignment

Authors: Wenjiang Zhang, Sifeng Wang, Yuwei Su, Xinyu Li, Chen Zhang, Suyu Zhong

Organizations: School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing, China · Queen Mary School Hainan, Beijing University of Posts and Telecommunications, Hainan 570100, China

Abstract

Visual neural decoding seeks to reconstruct or infer perceived visual stimuli from brain activity patterns, providing critical insights into human cognition and enabling transformative applications in brain-computer interfaces and artificial intelligence. Current approaches, however, remain constrained by the scarcity of high-quality stimulus-brain response pairs and the inherent semantic mismatch between neural representations and visual content. Inspired by perceptual variability and co-adaptive strategy of the biological systems, we propose a novel self-supervised architecture, named NeuroBridge, which integrates Cognitive Prior Augmentation (CPA) with Shared Semantic Projector (SSP) to promote effective cross-modality alignment. Specifically, CPA simulates perceptual variability by applying asymmetric, modality-specific transformations to both EEG signals and images, enhancing semantic diversity. Unlike previous approaches, SSP establishes a bidirectional alignment process through a co-adaptive strategy, which mutually aligns features from two modalities into a shared semantic space for effective cross-modal learning. NeuroBridge surpasses previous state-of-the-art methods under both intra-subject and inter-subject settings. In the intra-subject scenario, it achieves the improvements of 12.3% in top-1 accuracy and 10.2% in top-5 accuracy, reaching 63.2% and 89.9% respectively on a 200-way zero-shot retrieval task. Extensive experiments demonstrate the effectiveness, robustness, and scalability of the proposed framework for neural visual decoding.

Figures & tables

Figure 1: Illustration of human visual cognition and EEG variability. When viewing a visual stimulus, individuals focus on different semantic regions (e.g., parts of a cat), leading to varied EEG responses influenced by attention, mental state, and individual differences.

Figure 2: Overall framework of NeuroBridge. During training, cognitively augmented image-EEG pairs are encoded and projected into a shared semantic space using contrastive learning. At inference, EEG signals are decoded by matching their embeddings to visual concepts.

Intra-Subject: train and test on one subject
Method		Sub 1	Sub 2	Sub 3	Sub 4	Sub 5	Sub 6	Sub 7	Sub 8	Sub 9	Sub 10	Average
BraVL	Top-1	6.1	4.9	5.6	5.0	4.0	6.0	6.5	8.8	4.3	7.0	5.8
BraVL	Top-5	17.9	14.9	17.4	15.1	13.4	18.2	20.4	23.7	14.0	19.7	17.5
NICE	Top-1	13.2	13.5	14.5	20.6	10.1	16.5	17.0	22.9	15.4	17.4	16.1
NICE	Top-5	39.5	40.3	42.7	52.7	31.5	44.0	42.1	56.1	41.6	45.8	43.6
ATM	Top-1	25.6	22.0	25.0	31.4	12.9	21.3	30.5	38.8	34.4	29.1	27.1

Table 1: Overall accuracy (%) of 200-way zero-shot retrieval on THINGS-EEG2: Top-1 and Top-5.

Figure 3: Top-1 accuracy (%) for different transformation combinations.

Figure 4: Visual examples of applied image transformations.

# Transforms	1	2	3	4	5	6	7
Top-1 (%)	50.9	54.2	58.4	62.1	58.5	58.1	57.0
Top-5 (%)	81.4	84.5	88.1	89.8	88.1	87.1	86.0

Table 2: Top-1 and Top-5 accuracy (%) under different numbers of fused transformations.

Figure 5: Top-1 accuracy (%) across different projector types and feature dimensions.

Figure 6: Top-1 accuracy improvement (%) for various encoder architectures.

CPA		SSP	Top-1	Top-5
Image Prior	EEG Prior	SSP	Top-1	Top-5
✗	✗	✗	40.5	72.2
✓	✗	✗	60.0	89.1
✗	✓	✗	40.8	72.7
✗	✗	✓	41.5	73.5
✗	✓	✓	41.8	73.6

Table 3: Ablation study on the core components of NeuroBridge.

Figure 7: Semantic analysis on subject 8. (a) Similarity matrix. (b) Retrieved samples.

Appendix figures & tables13 assets

Supplementary material from the paper’s appendix.

Appendix

EEG Encoder	Channels	Parameters	Emb Dim
EEGNet	17	2.34M	1024
EEGNet	63	2.34M	1024
TSConv	17	2.56M	1024
TSConv	63	2.63M	1024
ATM	17	3.13M	1024
ATM	63	3.20M	1024

Appendix

Table 4: Details of different EEG encoders.

Image Encoder	Parameters	Emb Dim
RN50	38.32M	1024
RN101	56.26M	512
ViT-B-16	86.19M	512
ViT-B-32	87.85M	512
ViT-L-14	303.97M	768
ViT-H-14	632.08M	1024

Appendix

Table 5: Details of different Image encoders.

Intra-Subject: train and test on one subject
Method		Sub 1	Sub 2	Sub 3	Sub 4	Average
NICE	Top-1	9.6	18.5	14.2	9.0	12.8
NICE	Top-5	27.8	47.8	41.6	26.6	36.0
UBP	Top-1	15.0	46.0	27.3	18.5	26.7
UBP	Top-5	38.0	80.5	59.0	43.5	55.2
NeuroBridge (Ours)	Top-1	16.5	53.7	40.4	18.1	32.2

Appendix

Table 6: Overall accuracy (%) of 200-way zero-shot retrieval on THINGS-MEG: Top-1 and Top-5.

Method	$\ell_{2}$ -EEG	$\ell_{2}$ -Image	Top-1	Top-5
Plain	✗	✗	54.4	85.5
Sym	✓	✓	46.4	79.0
Inv-Asym	✓	✗	38.6	71.7
Asym	✗	✓	63.2	89.9

Appendix

Table 7: Comparison of

\ell_{2}

normalization strategies on EEG and image representations.

Figure 8: Top-1 accuracy (%) of Vanilla on the THINGS-EEG2 dataset.

Figure 9: Top-5 accuracy (%) of Vanilla on the THINGS-EEG2 dataset.

Figure 10: Top-1 accuracy (%) of NeuroBridge on the THINGS-EEG2 dataset.

Figure 11: Top-5 accuracy (%) of NeuroBridge on the THINGS-EEG2 dataset.

Figure 12: Top-1 accuracy improvement (%) on the THINGS-EEG2 dataset.

Figure 13: Top-5 accuracy improvement (%) on the THINGS-EEG2 dataset.

Batch Size	32	64	128	256	512	1024	2048
Top-1 (%)	54.6	54.9	57.6	59.4	60.8	63.2	62.2
Top-5 (%)	83.9	84.8	86.5	87.8	89.1	89.9	89.8

Appendix

Table 8: Top-1 and Top-5 accuracy (%) under different batch sizes.

Temperature	Learnable	0.001	0.005	0.01	0.05	0.1	0.5
Top-1 (%)	62.1	59.5	61.0	60.7	62.0	62.2	63.6
Top-5 (%)	90.2	88.9	89.0	89.2	89.8	89.6	91.0

Appendix

Table 9: Top-1 and Top-5 accuracy (%) under different temperatures.

EEG-Based Visual Decoding

EEG-based visual decoding maps brain signals recorded during visual perception to image identities, attributes, or reconstructed scenes.

See all 45 papers

MindAlign: Bridging EEG, Vision, and Language for Zero-Shot Visual Decoding
May 23, 2026Zexuan Chen, Sichao Liu, Runhao Lu +4EEG-Based Visual Decoding Cross-Modal Alignment
KTH, SWeden · University of Cambridge, UK · EPFL, Switzerland +2
STAMBRIDGE: Spectral-Temporal Amplitude-aware Mid-Feature Bridge for EEG Visual Decoding
May 22, 2026Jiahe Meng, Weiming Zeng, Yueyang Li +5Cross-Modal Knowledge Distillation Representation Learning
Lab of Digital Image and Intelligent Computation, Shanghai Maritime University, Shanghai 201306, China · Department of Language Science and Technology, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong SAR, China · Department of Neurology, Affiliated Lianyungang Hospital of Xuzhou Medical University, Lianyungang 222002, China +1
Neural Visual Decoding via Cognitive guided Adaptive Blurring and Information Constrained Alignment
May 14, 2026Fan Yin, Chuhang Zheng, Peiliang Gong +2EEG-Based Visual Decoding Cross-Modal Alignment
Department of Artificial Intelligence, Nanjing University of Aeronautics and Astronautics, Nanjing, China · Department of Electrical and Information Engineering, Tianjin University, Tianjin, China

May 23, 2026J/K move · Enter open · S save

MindAlign: Bridging EEG, Vision, and Language for Zero-Shot Visual Decoding

Zexuan Chen, Sichao Liu, Runhao Lu, Huichao Qi+3

KTH, SWeden · University of Cambridge, UK · EPFL, Switzerland+2

EEG-Based Visual Decoding Cross-Modal Alignment Multimodal Contrastive Learning

Open paper AskPDF arXiv

Visual decoding from brain signals is a key challenge at the intersection of computer vision and neuroscience, requiring methods that bridge neural representations and computational models of vision. We introduce a tri-modal contrastive framework for EEG-based visual decoding that aligns EEG, visual, and textual representations within a unified latent space. Our approach follows a two-stage design. First, we pre-train an EEG encoder via masked reconstruction on unlabeled trials, learning spatio-temporal regularities that transfer robustly to downstream tasks. Second, we jointly align EEG, image, and LLM-generated textual descriptions through contrastive learning, where text supervision acts as a semantic regularizer that injects linguistic structure into the shared space without overwhelming the primary EEG-image signal. The encoder integrates subject-specific adaptation, graph-attention over channels, and temporal-spatial convolutional embeddings. On the Things-EEG2 200-way zero-shot benchmark, our framework achieves 54.1% Top-1 and 83.4% Top-5 accuracy, substantially exceeding the strongest prior baseline (32.4% / 64.0%), with paired Wilcoxon tests confirming significance (p < 0.01) over all in-subject baselines. We validate generalization on Things-MEG. Analysis reveals that compact embedding geometries (CN-CLIP) outperform much larger backbones, and that decoding aligns with established neurophysiology of visual processing. This work is a critical step towards robust, semantically-grounded visual decoding from non-invasive temporal neural signals. The source code is publicly available in https://github.com/anon-eeg/eeg_image_decoding.

Abstract

EEG-Based Visual Decoding

Explore similar work

MindAlign: Bridging EEG, Vision, and Language for Zero-Shot Visual Decoding

STAMBRIDGE: Spectral-Temporal Amplitude-aware Mid-Feature Bridge for EEG Visual Decoding

Neural Visual Decoding via Cognitive guided Adaptive Blurring and Information Constrained Alignment