GlycoEMG Open Dataset

This is the official repository for the GlycoEMG Dataset, a comprehensive open-source database combining Surface Electromyography (sEMG) signals and physiological Glucose measurements.

Inspired by clinical ML datasets like PhysioNet and ECG-QA, this repository aims to bridge the gap between muscular fatigue and glucose level fluctuations to train machine learning models and predictive AIs.

---

Dataset Snapshot

Here is an example of what the raw Electromyography (EMG) signals look like, intrinsically linked to clinical metadata such as Glucose Drop (Delta):

The dataset includes high-frequency sampling at 250Hz and 500Hz, capturing detailed muscular fiber recruitment.

---

Data Acquisition Protocol

The data was meticulously gathered to ensure consistency across all sessions, using the following clinical pipeline:

1. Dietary Intake & Wait Period:
   • A pre-workout meal was provided to all subjects.
   • A waiting period was enforced to reach the glucose peak: 45 minutes for the control group and 2 hours for the diabetes group.
2. Pre-Exercise Glucose: Blood glucose levels were measured immediately prior to the exercise.
3. Hardware Setup: sEMG sensors were placed on the Quadriceps (Vastus lateralis). The raw signal was transmitted in real-time via an ESP32 microcontroller using a Bluetooth (BLE) protocol to our custom desktop application.
4. Exercise Execution: Subjects performed free squats carrying approximately 30% of their body weight, pushing close to muscular failure.
5. Post-Exercise Glucose: A final blood glucose measurement was taken immediately after the exercise concluded to capture the glucose delta.

---

File Structure

To guarantee privacy and facilitate Machine Learning research, the data is distributed with all PII (Personally Identifiable Information) removed. The repository is structured as follows:

glycoemg-dataset/
├── data/
│   └── quick_access/               # Ready-to-use extracted data
│       ├── csv/                    # Time-series data (pandas friendly)
│       │   ├── session_001.csv
│       │   └── ...
│       └── json/                   # Hierarchical data (Metadata + Array)
│           ├── session_001.json
│           └── ...
├── glycoemg/                       # Python Package for data manipulation
│   ├── data.py                     # Data loaders (HDF5 & CSV)
│   ├── metrics.py                  # Fatigue metrics (RMS, MDF)
│   └── viz.py                      # Visualization utilities
├── notebooks/                      # Jupyter Notebooks Demonstrations
│   ├── 01_data_loading.ipynb
│   └── 02_emg_glucose_correlation.ipynb
├── scripts/                        # Utility scripts (Split, BIDS export)
└── environment.yml                 # Conda environment definition

---

Demonstrations

We provide Jupyter Notebooks to facilitate users in skimming over the dataset and extracting clinical insights:

• 01 Data Loading: A step-by-step guide to downloading the anonymized database and loading a session using the glycoemg python package.
• 02 EMG & Glucose Correlation: Extraction of the Median Frequency (MDF) of the signal and its correlation with the Glucose Delta.

---

Setup & Installation

You can install the glycoemg package locally to interact with the raw signals programmatically.

# Clone the repository
git clone https://github.com/Gonzalo-A-G/glycoemg-dataset.git
cd glycoemg-dataset

# Create and activate Conda environment
conda env create -f environment.yml
conda activate glycoemg

# Install the package in editable mode
pip install -e .

To fetch the raw large HDF5 files (not hosted on GitHub due to size limitations), run:

bash scripts/download_dataset.sh

---

License

This dataset and code are distributed under the MIT License.