OCT Dataset: Ex-vivo Open-Sky Porcine Eye Imaging

This repository contains a dataset of Optical Coherence Tomography (OCT) images collected from ex-vivo open-sky and intact porcine eyes during robot assisted subretinal injections. The images have been hand-labeled to identify three key surfaces:

1. Needle (1)
2. Internal Limiting Membrane (ILM) (2)
3. Retinal Pigment Epithelium (RPE) (3)

The dataset is intended for use in developing models for segmentation or other types of image analysis. There is a total of 1560 images.

Dataset Overview

A 42 gauge (100 micron thickness) needle was used during the procedures.
The images in this dataset are annotated using the Supervisely platform. Each pixel in the labeled images belongs to one of the following classes:

• 0: Background
• 1: Needle
• 2: Internal Limiting Membrane (ILM)
• 3: Retinal Pigment Epithelium (RPE)

Important Notes

• Because the bevel is difficult to accurately detect while the needle is deeper in the tissue the bevel is not annotated.
• The labeled images are not directly human-viewable since they contain discrete class values (0, 1, 2, and 3).
• For visualization purposes, we provide an example Jupyter notebook that demonstrates how to convert these labeled images into a human-viewable format using color mapping.
• Files labelled xyz-intact are collected from intact porcine eyes instead of open-sky.

Repository Structure

.
├── data/
│   ├── jul9-3/
│   │   ├── ann/                 # JSON files for Supervisely annotations
│   │   ├── img/                 # Original OCT images
│   │   ├── masks_machine/       # Machine-readable masks (pixel values: 0, 1, 2, 3)
│   ├── jul9-4/
│   │   ├── ann/                 # JSON files for Supervisely annotations
│   │   ├── img/                 # Original OCT images
│   │   ├── masks_machine/       # Machine-readable masks (pixel values: 0, 1, 2, 3)
│   ├── ...
│   ├── meta.json                # Metadata for supervisely
│   ├── obj_class_to_machine_color.json  # Metadata for supervisely
├── notebooks/
│   ├── visualize_labels.ipynb    # Notebook for visualizing the annotations
│ 
├── README.md

Example Notebooks

1. Visualizing the Annotations: In the visualize_labels.ipynb notebook, you can find a function to convert the label images into a human-viewable format.

Installation and Usage

To work with the dataset and provided notebook, you'll need to install the required Python packages. The basic setup is as follows:

Requirements

• Python 3.12 (Probably works with any version of Python 3)
• numpy
• matplotlib
• opencv-python (for image processing)
• jupyterlab or notebook (for running the example notebooks)

Installation

1. Clone the repository:

   git clone [email protected]:your-user-name/subretinal-injection-oct-dataset.git
   cd subretinal-injection-oct-dataset

2. Install dependencies:

   pip install numpy
   pip install matplotlib
   pip install opencv-python
   pip install jupyterlab

3. Run the example notebooks:

   jupyter notebook notebooks/visualize_labels.ipynb

Dataset Citation

If you use this dataset in your research or projects, please cite it appropriately.

TODO: add citation

License

This dataset is provided under the MIT License.