🧠 MEG Signal Decomposition Project: VMD and MVMD Analysis

🎯 Project Overview

This repository provides a complete pipeline for analyzing Magnetoencephalography (MEG) resting-state data using several signal decomposition methods:

• Empirical Mode Decomposition (EMD)
• Univariate Variational Mode Decomposition (VMD)
• Multivariate Variational Mode Decomposition (MVMD)

The primary objective of this project is to investigate whether extracting narrowband oscillatory modes can improve Multifractal Analysis (MFA), particularly for identifying scale-invariant properties in neural dynamics.

The MFA procedures rely on the pymultifracs library, developed within the Mind Team (CEA NeuroSpin) by Merlin Dumeur.

---

🧪 Decomposition Rationale

This project evaluates three decomposition strategies:

• EMD: Due to its lack of rigorous mathematical foundations, EMD was used only on simulated data and was not retained for real MEG analysis.

• VMD: VMD demonstrated strong performance and was applied to both simulated and real-world data.

• MVMD: As the multivariate extension of VMD, MVMD aims to identify shared oscillatory modes across multiple MEG sensors, potentially capturing spatial interactions and reducing the need for channel-wise decomposition.

---

👥 How to Use This Repository

This project supports two main use cases:

1. Reproduce the Internship Results Follow the installation steps and run the notebooks in order to replicate the full analysis pipeline.

2. Apply EMD, VMD, or MVMD to Your Own Dataset Use the simplified starting notebooks to quickly test these decomposition methods on custom time-series data.

---

🚀 Getting Started

1. 🐍 Setting up the Conda Environment

We recommend creating the environment via Conda:

# Create the environment from the provided requirements.txt
conda env create -f requirements.txt
conda activate meg-vmd

---

2. 📁 Data Acquisition (Important)

⚠️ The required data files are NOT included in this repository.

To run the main analysis notebooks, please contact the project maintainers to obtain the necessary files. Once acquired, place them in the following directory structure:

Required File	Description	Destination Folder
sub-01_ses-01_task-rest_proc-filt_raw.fif	Raw MEG recording used for VMD and MVMD pipelines	data/
mvmd_modes_sub-01.npz	Pre-computed MVMD results for analysis	results/real/MVMD/modes/

---

💻 Option 1: Reproducing the Report Results

After activating the environment and placing the data in the appropriate folders, start Jupyter Lab and run the notebooks in order:

Notebook	Description	Dependencies
00_simulation_pipeline.ipynb	Synthetic data benchmark	None
01_vmd_pipeline.ipynb	Univariate VMD on real MEG	data/sub-01...raw.fif
02_mvmd_on_realdata_pipeline.ipynb	Analysis of MVMD results	results/real/MVMD/modes/

---

🛠️ Running Your Own MVMD Decomposition if you dont want to use my result

MVMD decomposition is computationally demanding and requires a GPU.

If you want to generate your own MVMD modes:

1. Navigate to the GPU-dedicated notebook:

   cd notebooks/notebooks_gpu/

2. Run the notebook using a machine with GPU support (Google Colab Pro, a CUDA-equipped workstation, or a server).

3. The notebook will output a file such as:

   mvmd_modes_sub-01.npz
   

   Place this file in:

   results/real/MVMD/modes/
   

   and rerun the analysis notebook (02_mvmd_on_realdata_pipeline.ipynb).

---

💻 Option 2: Applying Methods to Your Own Dataset

To quickly test EMD, VMD, or MVMD on your own dataset, use the starter notebooks:

• Path: notebooks/extras/starting_notebooks/
• Content: These notebooks contain simplified simulation examples and minimal workflows that can be easily adapted to custom time-series.

---

📬 Contact

For additional information:

• Pipeline, notebooks, and analysis workflow: Sabrine Bendimerad

• Access to the MEG data used in this project: Philippe Ciuciu

• Multifractal Analysis (MFA) and the pymultifracs library: Merlin Dumeur