CAMEL-CLIP: Channel-aware Multimodal Electroencephalography-text Alignment for Generalizable Brain Foundation Models

Hanseul Choi; Jinyeong Park; Seongwon Jin; Sungho Park; Jibum Kim

cs.LGFeb 27, 2026

CAMEL-CLIP: Channel-aware Multimodal Electroencephalography-text Alignment for Generalizable Brain Foundation Models

Authors: Hanseul Choi, Jinyeong Park, Seongwon Jin, Sungho Park, Jibum Kim

Organizations: Department of Computer Science and Engineering, Incheon National University, Yeonsu-gu, 22012, Incheon, Republic of Korea · Department of Artificial Intelligence, Inha University, Michuhol-gu, 22212, Incheon, Republic of Korea · Center for Brain-Machine Interface, Incheon National University, Yeonsu-gu, 22012, Incheon, Republic of Korea

Abstract

Electroencephalography (EEG) foundation models have shown promise for learning generalizable representations, yet they remain sensitive to channel heterogeneity, such as changes in channel composition or ordering. We propose channel-aware multimodal EEG-text alignment contrastive language-image pretraining (CAMEL-CLIP), a contrastive EEG-text multimodal foundation model designed to be robust to heterogeneous channel configurations and widely applicable to diverse downstream tasks. CAMEL-CLIP introduces three key components: (1) channel attribute-based positional encoding, which identifies channels through semantic information; (2) dynamic channel projection, which generates variable-length embeddings by independently projecting each channel without feature compression; and (3) dual-level contrastive learning, which jointly performs channel-level and sample-level contrastive learning to capture both channel-specific and global signal characteristics. Experimental results demonstrate that CAMEL-CLIP achieves state-of-the-art performance under linear-probing and outperforms existing foundation models that rely on full-finetuning.

Figures & tables

Figure 1: UMAP visualization of embedding vectors for each model. (a) EEG-CLIP embeddings. (b) CAMEL-CLIP channel-wise embeddings. Unlike the baseline model (EEG-CLIP), whose embeddings are not separated by channel, CAMEL-CLIP forms distinct embedding distributions for different channels.

Figure 2: Paradigm shift in deep learning-based EEG decoding. (a) Conventional task-specific models, where a separate EEG encoder is trained for each task and dataset. (b) Prior brain foundation models, which enable multi-task transfer via pretrained weights but still require finetuning on downstream datasets due to channel heterogeneity. (c) CAMEL-CLIP, which mitigates channel heterogeneity and supports multiple tasks with linear-probing alone.

Figure 3: Framework of the proposed model. (a) Channel attribute-based positional encoding. (b) Dynamic channel projection for channel-wise embeddings. (c) dual-level contrastive learning at the channel and sample levels. Here, A1, FP2, P4, FP3, and T6 annotate example channel names.

Figure 4: Pipeline of proposed synthetic report ensemble prompt generation. (a) The proposed text prompt generation method for text-based classification generates multiple synthetic reports and averages their embeddings. (b) An example of generated synthetic report.

Figure 5: Cosine similarity distributions across prompting strategies. Cosine similarity between prompts for normal labels in the pathological task and text embeddings from the validation set. The prompts from the proposed ensemble method show a higher mean cosine similarity.

$p_{remove}$	Balanced accuracy
0	0.7330
0.25	0.7325
0.50	0.7442
0.75	0.7230

Table 1: Average text-based classification balanced accuracy across varying dynamic channel removal probability

p_{remove}

. This table reports the mean balanced accuracy of text-based classification on the TUAB validation set for different removal probabilities.

n	Balanced accuracy
1	0.7387
3	0.7442
5	0.7416

Table 2: Average text-based classification balanced accuracy across CCL sampling values of

n

. This table reports the mean balanced accuracy of text-based classification on the TUAB validation set for different CCL sampling values of

n

Methods	pathological	gender	age
EEG-CLIP	0.4467	0.5360	0.4395
Proposed	0.5640	0.6437	0.6751

Table 4: Text-based classification (single prompt). This table reports text-based classification performance on the three TUAB tasks using a single prompt.

Methods	pathological	gender	age
EEG-CLIP	0.7905	0.5608	0.6718
Proposed	0.8158	0.6463	0.7125

Table 5: Text-based classification (ensemble prompt). This table reports text-based classification performance on the three TUAB tasks using ensemble prompts.

Figure 6: Channel-wise EEG embedding visualizations on the TUAB dataset. (a) Channel-wise EEG embeddings from the model trained without channel-level contrastive learning (CCL). (b) Channel-wise EEG embeddings from the model trained with CCL. (c) Channel-wise EEG embeddings for five channels ( i.e., F4, FZ, O1, P3, and T3) from the model trained using all channels. (d) Channel-wise EEG embeddings from the model trained excluding the five channels.

Figure 7: Channel-wise EEG embedding visualizations on the CHB-MIT dataset. (a) Channel-wise EEG embeddings from the model trained without channel-level contrastive learning (CCL). (b) Channel-wise EEG embeddings from the model trained with CCL.

Figure 8: Visualization of the mean cosine similarity between channel embeddings. We evaluate the similarity for two channels ( i.e., F4 and FZ) using the model trained while excluding five channels ( i.e., F4, FZ, O1, P3, and T3). (a) Mean cosine similarity between the F4 embedding and the embeddings of other channels. (b) Mean cosine similarity between the FZ embedding and the embeddings of other channels.

Methods	pathological	gender	age
EEG-CLIP	0.8148 $\pm$ 0.0026	0.5798 $\pm$ 0.0018	0.6840 $\pm$ 0.0003
Proposed	0.8356 $\pm$ 0.0011	0.6747 $\pm$ 0.0020	0.7348 $\pm$ 0.0024

Table 6: TUAB dataset classification with a trained classifier head. This table reports linear-probing performance on the three TUAB tasks.

Methods	Balanced accuracy	AUC-PR
EEGNet	0.5658 $\pm$ 0.0106	0.1914 $\pm$ 0.0182
FFCL	0.6262 $\pm$ 0.0104	0.2049 $\pm$ 0.0346
BIOT	0.7068 $\pm$ 0.0457	0.3277 $\pm$ 0.0460
LaBraM-Base	0.7075 $\pm$ 0.0358	0.3287 $\pm$ 0.0402
EEG-CLIP	0.6843 $\pm$ 0.0601	0.2191 $\pm$ 0.0818
EEG-CLIP (LP)	0.5000 $\pm$ 0.0000	0.0981 $\pm$ 0.0233

Table 7: CHB-MIT dataset classification with a trained classifier head. This table reports performance on seizure detection on the CHB-MIT dataset.

Methods	Balanced accuracy	AUC-PR
EEGNet	0.9232 $\pm$ 0.0104	0.9639 $\pm$ 0.0093
FFCL	0.9314 $\pm$ 0.0038	0.9717 $\pm$ 0.0021
BIOT	0.9358 $\pm$ 0.0052	0.9736 $\pm$ 0.0034
LaBraM-Base	0.9409 $\pm$ 0.0079	0.9798 $\pm$ 0.0093
EEG-CLIP	0.6582 $\pm$ 0.0095	0.6549 $\pm$ 0.0127
EEG-CLIP (LP)	0.5030 $\pm$ 0.0046	0.5548 $\pm$ 0.1032

Table 8: MUMTAZ-2016 dataset classification. This table reports performance on major depressive disorder classification on the MUMTAZ-2016 dataset, where LP denotes the linear-probing setting.

Methods	pathological	gender	age
w/o CAPE, DCP, DCL	0.7270	0.6633	0.4874
w/o DCP, DCL	0.7172	0.6545	0.4407
w/o DCL	0.8172	0.6300	0.6661
Proposed	0.8158	0.6463	0.7125

Table 9: Ablation study on TUAB. This table summarizes text-based classification performance on the TUAB dataset, where CAPE, DCP, and DCL denote channel attribute-based positional encoding, dynamic channel projection, and dual-level contrastive learning, respectively.

Methods	Balanced accuracy	AUC-PR
w/o CAPE, DCP, DCL	0.7888 $\pm$ 0.0075	0.2970 $\pm$ 0.0059
w/o DCP, DCL	0.8002 $\pm$ 0.0071	0.2425 $\pm$ 0.0070
w/o DCL	0.7544 $\pm$ 0.0048	0.2576 $\pm$ 0.0064
Proposed	0.7988 $\pm$ 0.0046	0.3689 $\pm$ 0.0064

Table 10: Ablation study on CHB-MIT. This table summarizes text-based classification performance on the CHB-MIT dataset, where CAPE, DCP, and DCL denote channel attribute-based positional encoding, dynamic channel projection, and dual-level contrastive learning, respectively. All models are evaluated under the linear-probing setting.

Methods	Balanced accuracy	AUC-PR
w/o CAPE, DCP, DCL	0.8936 $\pm$ 0.0006	0.9610 $\pm$ 0.0003
w/o DCP, DCL	0.9161 $\pm$ 0.0015	0.9807 $\pm$ 0.0004
w/o DCL	0.9300 $\pm$ 0.0009	0.9836 $\pm$ 0.0003
Proposed	0.9439 $\pm$ 0.0019	0.9900 $\pm$ 0.0001

Table 11: Ablation study on MUMTAZ-2016. This table summarizes text-based classification performance on the MUMTAZ-2016 dataset, where CAPE, DCP, and DCL denote channel attribute-based positional encoding, dynamic channel projection, and dual-level contrastive learning, respectively. All models are evaluated under the linear-probing setting.

Figure 9: Sample of EEG-text retrieval from the TUAB dataset. (a) Text report of query EEG sample. (b) Retrieval report.

Methods	pathological	gender	age
EEG-CLIP @ 1	0.7994	0.5273	0.5924
EEG-CLIP @ 5	0.7752	0.5316	0.5875
EEG-CLIP @ 10	0.7733	0.5331	0.5899
EEG-CLIP @ 100	0.7731	0.5319	0.5940
Proposed @ 1	0.8182	0.6316	0.6938
Proposed @ 5	0.8136	0.6394	0.6712

Table 12: Precision on the retrieval task. This table reports retrieval precision on the three TUAB tasks. Precision is computed as the fraction of retrieved reports within the top-

K

results whose labels match the ground-truth label.

EEG Foundation Models

Electroencephalography (EEG) foundation models learn reusable representations of brain signals for transfer across people, tasks, and recording setups.

See all 96 papers

Standing on the Shoulders of Giants: Rethinking EEG Foundation Model Pretraining via Multi-Teacher Distillation
Mar 4, 2026Chenqi Li, Yu Liu, Shuo Zhang +2EEG Foundation Models Multi-Teacher Knowledge Distillation
Department of Engineering Science, University of Oxford.
BAT-CLIP: Trimodal Alignment of Brain, Audio and Text
Sep 25, 2026Suhyun Kim, Jinmo Han, Danny Dongyeop Han +9Cross-Modal Alignment Multimodal Embedding
Yonsei University, Seoul, Republic of Korea · Seoul National University, Seoul, Republic of Korea · Dartmouth College, Hanover, NH, USA
DEFT: Disentanglement-Enhanced Fine-Tuning for EEG Foundation Models
Oct 10, 2026Yuqiu Deng, Muyun Jiang, Zongpeng Zhang +8EEG Foundation Models Cross-Subject EEG Decoding
Uploading Inc · College of Computing and Data Science, Nanyang Technological University · Department of Biostatistics, School of Public Health, Peking University +2

Mar 4, 2026J/K move · Enter open · S save

Standing on the Shoulders of Giants: Rethinking EEG Foundation Model Pretraining via Multi-Teacher Distillation

Chenqi Li, Yu Liu, Shuo Zhang, Timothy Denison+1

Department of Engineering Science, University of Oxford.

EEG Foundation Models Multi-Teacher Knowledge Distillation Cross-Modal Knowledge Distillation

Open paper AskPDF arXiv

Pretraining for electroencephalogram (EEG) foundation models has predominantly relied on self-supervised masked reconstruction, a paradigm largely adapted from and inspired by the success of vision and language foundation models. However, unlike images and text, EEG datasets are notoriously expensive to collect and characterized by low signal-to-noise ratio. These challenges introduce difficulties in scaling the EEG foundation models and capturing the underlying neural semantics through reconstruction. In this work, we ask the question: can we stand on the shoulders of well-established foundation models from well-represented modalities to bootstrap the pretraining of EEG foundation models? We first demonstrate that mainstream foundation models, such as those from vision and time series, transfer surprisingly well to EEG domain. To this end, we propose the Multi-Teacher Distillation Pretraining (MTDP) framework for pretraining EEG foundation models via a two-stage multi-teacher distillation. In the first stage, we introduce a learnable gating network to fuse representations from diverse teachers (e.g., DINOv3 and Chronos) via a masked latent denoising objective. In the second stage, we distill the fused representation into an EEG foundation model. Extensive evaluations across 9 downstream tasks and 12 datasets demonstrate that our MTDP-based EEG foundation model outperforms its self-supervised counterparts while requiring only 25% of the pretraining data.

Abstract

EEG Foundation Models

Explore similar work

Standing on the Shoulders of Giants: Rethinking EEG Foundation Model Pretraining via Multi-Teacher Distillation

BAT-CLIP: Trimodal Alignment of Brain, Audio and Text

DEFT: Disentanglement-Enhanced Fine-Tuning for EEG Foundation Models