ecg-image-and-signal-dataset

KO | EN

gitlite — search

#hacktoberfest #ai #typescript #android #machine-learning #rust #deep-learning #linux #docker #macos #java #vue #pytorch #security

ecg-image-and-signal-dataset

(★ 12)

A public ECG dataset featuring raw ECG images, per-lead segmentations, and digitized signals (PNG, BMP & JSON) for lead detection and separation.

ecg-image-and-signal-dataset Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# ECG Image and Signal Dataset Generator

This repository provides an open-source Python framework for generating customizable, large-scale synthetic ECG images from signals. Multiple datasets are included to support deep learning-based research on ECG digitization, lead and lead-name detection, and waveform segmentation.

The dataset for waveform segmentation is provided in two versions: normal and overlapping. In the overlapping version, signals from adjacent leads (above or below) are superimposed onto a target lead. At the same time, the corresponding masks remain clean, containing only the true waveform of the target lead.

More information on our paper.
- Paper at Journal of Medical Signals & Sensors [SynthECG: Python Framework and ECG Image Datasets for Digitization, Lead Detection, and Waveform Segmentation](https://journals.lww.com/jmss/fulltext/2026/03300/synthecg__python_framework_and_ecg_image_datasets.3.aspx)
- Datasets at Zenodo: [https://doi.org/10.5281/zenodo.15484519](https://doi.org/10.5281/zenodo.15484519)
- Open-source code at GitHub: [https://github.com/rezakarbasi/ecg-image-and-signal-dataset](https://github.com/rezakarbasi/ecg-image-and-signal-dataset)

---

## 🧠 Citation
If you use this dataset or code, please cite both the dataset and the paper:

**Paper:**  
> Rahimi, Masoud*; Karbasi, Reza*; Vahabie, Abdol-Hossein. SynthECG: Python Framework and ECG Image Datasets for Digitization, Lead Detection, and Waveform Segmentation. Journal of Medical Signals & Sensors 16(3):8, March 2026. | DOI: 10.4103/jmss.jmss_58_25

**Dataset:**
> Rahimi, M., Karbasi, R., & Vahabie, A. H. (2025). *An Open-Source Python Framework and Synthetic ECG Image Datasets for Digitization, Lead and Lead Name Detection, and Overlapping Signal Segmentation*. University of Tehran.

---

## 📦 Features

- Generate realistic ECG print-style images in various layouts: `3x1`, `3x4`, `6x2`, `12x1` (in `3x4` and `6x2`, lead II can be printed in full at the bottom)
- Annotate images for:
  - **Lead Region Detection** (YOLO format)
  - **Lead Name Detection** (YOLO format)
  - **Pixel-wise Segmentation** (with both normal and overlapping signals for U-Net)
- Compatible with deep learning models like YOLO and U-Net
- Includes paired image, mask, and time-series data
- Overlapping segmentation support: clean masks for superimposed leads

---

## 📊 Datasets Provided

| Dataset Task               | Format/Annotations                              | Sample Size |
|----------------------------|-------------------------------------------------|-------------|
| ECG Digitization           | ECG images + ground truth signals               | 2,000       |
| Lead & Lead Name Detection | Bounding boxes (YOLO format)                    | 2,000       |
| Segmentation (Normal)      | Cropped leads + masks + ground truth signals    | 20,000      |
| Segmentation (Overlapping) | Overlapping leads + clean masks                 | 102         |

### Dataset Samples

<figure>
 <img src="images/figure 1.jpg" width="600">
 <figcaption>Figure 1: Lead & Lead Name Detection dataset: ECG images annotated with bounding boxes for lead regions and lead names. The color of each box indicates the class ID. Class 0 corresponds to lead waveform regions, whereas classes 1 to 12 represent the lead names I, II, III, aVR, aVL, aVF, and V1–V6, respectively. The bounding boxes for lead names are notably smaller than those for the waveform regions. </figcaption>
 </figure>

<figure>
 <img src="images/figure 2.jpg" width="600">
 <figcaption>Figure 2: Illustration of the YOLO bounding box format, which includes the class ID, normalized center coordinates (x, y), and the bounding box width and height (w, h), all expressed relative to the image dimensions. </figcaption>
 </figure>

<figure>
 <img src="images/figure 3.jpg" width="600">
 <figcaption>Figure 3: Normal and overlapping segmentation dataset: Top – Single-lead ECG image with its corresponding mask shown on the right. Middle and bottom – Overlapping single-lead ECG images with clean masks displayed on the right. masks are in U-Net format</figcaption>
 </figure>

---

## ⚙️ Usage

### 1. Install Dependencies
```bash
pip install -r requirements.txt
```

### 2. Prepare Raw Data
Download the raw WFDB files from [PTB-XL](https://physionet.org/content/ptb-xl/1.0.3/), unzip them, and place the contents into the `./dataset/signal_dataset/` directory.

### 3. Run the Dataset Generator
```bash
python make_dataset.py \
    --sample_each_lead_train 5 \
    --sample_each_lead_val 2 \
    --sample_each_lead_test 2 \
    --row_height 3
```

Additional arguments available (see `make_dataset.py` for full list):
- `--signal_dataset_path` 
- `--datasets_path`
- `--horizontal_scale`, `--vertical_scale`
- `--show_grid`, `--show_lead_name`, `--fontsize`

### 4. Output Directory
```
.
├── detection-dataset
│   ├── images
│   │   ├── 0_3by1.jpg
│   │   ├── ...
│   └── labels
│       ├── 0_3by1.txt
│       └── ...
├── digitization-dataset
│   ├── 0_3by1
│   │   ├── 0_3by1.jpg
│   │   ├── 0_3by1_lead_0.json
│   │   ├── 0_3by1_lead_1.json
│   │   └── 0_3by1_lead_2.json
│   └── ...
├── overlap-dataset
│   ├── with-overlap
│   │   ├── image
│   │   │   ├── 699_3by4_lead_10.bmp
│   │   │   └── ...
│       ├── mask-bmp
│   │   │   ├── 699_3by4_lead_10.bmp
│   │   │   └── ...
│   │   ├── mask-png
│   │   │   ├── 699_3by4_lead_10.png
│   │   │   └── ...
│   │   └── signal-json
│   │       ├── 699_3by4_lead_10.json
│   │       └── ...
│   └── without-overlap
│       ├── image
│       │   ├── 699_3by4_lead_0.png
│       │   └── ...
│       ├── mask-bmp
│       │   ├── 699_3by4_lead_0.bmp
│       │   └── ...
│       ├── mask-png
│       │   ├── 699_3by4_lead_0.png
│       │   └── ...
│       └── signal-json
│           ├── 699_3by4_lead_0.json
│           └── ...
├── segmentation
│   ├── image
│   │   ├── 0_3by1_lead_0.png
│   │   └── ...
│   ├── mask-bmp
│   │   ├── 0_3by1_lead_0.bmp
│   │   └── ...
│   ├── mask-png
│   │   ├── 0_3by1_lead_0.png
│   │   └── ...
│   └── signal-json
│       ├── 0_3by1_lead_0.json
│       └── ...
└── logs.json

```

---

## 📬 Contact
Feel free to open an issue or discussion if you need help.

For questions, please contact:
- Masoud Rahimi: mr.rahimi39@ut.ac.ir
- Reza Karbasi: arzkarbasi@gmail.com

ecg-image-and-signal-dataset

ecg-image-and-signal-dataset Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

ecg-image-and-signal-dataset

ecg-image-and-signal-dataset Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README