metis-brain-signal-foundation-model

METIS Brain Signal Foundation Model

METIS is a brain-signal foundation model for EEG/iEEG analysis. It supports signal representation learning, Signal-QA, zero-shot classification, and downstream signal-only classification fine-tuning.

Model Files

Quick Start

Download the checkpoint and place it in your project as:

checkpoints/metis.pt

For the zero-shot demo and example code, see the GitHub repository:

mingzhi-c/metis-brain-signal-foundation-model

Signal-Only Classification Fine-Tuning

import torch

from METIS import MetisClassifier, MetisConfig

config = MetisConfig(num_classes=5)
model = MetisClassifier(config)

checkpoint = torch.load("checkpoints/metis.pt", map_location="cpu")
model.backbone.load_state_dict(checkpoint, strict=False)

x = torch.randn(1, 10, 1000)
y = model(x)

print(y.shape)  # torch.Size([1, 5])

Intended Use

• Research on EEG/iEEG foundation models
• Brain-signal representation learning
• Zero-shot and few-shot brain-signal analysis
• Downstream signal classification fine-tuning

Limitations

METIS is a research model and is not a standalone clinical diagnostic system. Downstream use requires task-specific validation, appropriate preprocessing, and expert review.

More Resources

Additional demo data, downstream examples, and documentation will be released progressively in the GitHub repository.