README.md

metadata

license: apache-2.0
language:
  - en
tags:
  - fMRI
  - EEG
  - neuroscience
  - brain-imaging
  - science
  - huggingscience
size_categories:
  - 100K<n<1M

📚 Dataset Card for CineBrain

CineBrain is a large-scale multimodal brain dataset comprising fMRI, EEG, and ECG recordings collected while participants watched episodes of The Big Bang Theory. It supports research on neural decoding, multimodal learning, and modality transfer in naturalistic narrative processing.

---

🧠 Dataset Description

Summary

• Participants: 6 subjects
• Stimuli: 30 episodes of The Big Bang Theory (first 18 minutes per episode)
• Recording time: 6 hours per subject (36 hours total)
• Modalities:
  • fMRI: TR = 0.8s
  • EEG: 64 channels, 1000 Hz
  • ECG: synchronous recording

Supported Tasks

• Multimodal Brain Analysis: Investigating relationships between audiovisual stimuli and neural responses
• Neural Decoding: Inferring cognitive states from fMRI and EEG signals
• Cross-Modal Learning: Learning shared representations across fMRI, EEG, and ECG
• Modality Transfer: Predicting fMRI from EEG and EEG from fMRI

---

📂 Dataset Structure

Repository Contents

• videos.tar: Video stimuli (8100 clips from 30 episodes)
• sub-00xx/: Participant folders with raw + preprocessed fMRI/EEG
• captions-qwen-2.5-vl-7b.json: Auto-generated video captions

Inside Each Participant Folder

• fMRI_raw_data.tar – raw fMRI
• fMRI_preprocessed_data.tar – preprocessed fMRI
• EEG_preprocessed_data.tar – preprocessed EEG

Data Statistics

Modality	Sampling	Duration	Size (approx.)
fMRI	TR=0.8s	6h/subject	~12 GB total
EEG	1000 Hz, 64 ch	6h/subject	~72 GB total
Video	30 eps × 18 min	8100 clips	~2.59 GB

Data Splits

• Subjects 1, 2, 6: Episodes 1–20 (5400 clips)
• Subjects 3, 4, 5: Episodes 1–10 and 21–30 (5400 clips)

Total: 36 hours of brain recordings across all subjects

---

📌 Important Notes

• Data Release: Fully open and downloadable.
• Cross-Dataset Correspondence: Subjects 1, 2, 3, and 4 in CineBrain correspond to Subjects 6, 8, 1, and 4 in the fMRI-Shape and fMRI-Objaverse datasets.

---

🏗 Dataset Creation

Motivation

CineBrain was designed to support naturalistic neuroscience research, focusing on:

• Narrative comprehension
• Multisensory integration
• Individual variability in brain responses
• Temporal dynamics of engagement

Source Data & Preprocessing

• fMRI: High temporal resolution (TR=0.8s)
• EEG: High sampling rate (1000 Hz)
• Preprocessing: Standard pipelines, artifact removal, quality control

⚠️ Ethics: All data anonymized. Please follow ethical guidelines when using human neuroimaging data.

---

⚖️ Considerations for Using the Data

Social Impact

This dataset may advance:

• Understanding of complex narrative processing
• Brain-computer interfaces
• Clinical applications for attention & comprehension disorders

Potential Biases

• Demographic bias: Limited participant diversity
• Cultural bias: English-language sitcom
• Selection bias: Likely university volunteers

---

📎 Additional Information

• License: Apache-2.0
• Languages: English audiovisual content

Citation Information

If you find our paper useful for your research and applications, please cite using this BibTeX:

@misc{gao2025cinebrain,
      title={CineBrain: A Large-Scale Multi-Modal Brain Dataset During Naturalistic Audiovisual Narrative Processing}, 
      author={Jianxiong Gao and Yichang Liu and Baofeng Yang and Jianfeng Feng and Yanwei Fu},
      year={2025},
      eprint={2503.06940},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2503.06940}, 
}