This repository contains a simple demonstration of how t-SNE can be used to represents different tissue types in histopathological images.
For this demonstration I used Jupyter Notebook to make it easy for others to reproduce. I also used the following Python packages:
- TensorBoard (with Embedding Projector plugin)
- medmnist (to access the MedMNIST histopathology image dataset)
- Pillow
- NumPy
To recreate a similar video to the one above, you can simply run the Jupyter Notebook (see under assets/).
Note that you may need to run the last cell a few times before the TensorBoard dashboard shows.
When it opens, enable Spherizie Data by clicking the checkbox on the left-hand sidebar.
Then click the T-SNE option underneath. This should start the simulation immediantly.
Then just play around with hyperparameters to get a simulation you are pleased with.
For the actual recording, I used screen recording on my macOS 14.6 using CMD + SHIFT + 5
and selecting a window, but use whichever tool you prefer :]
To convert recording to GIF, use any video converter like ezgif or similar.
If you use this illustration in a presentation or publication, please cite it using the repository URL.
Implementations were based on the code provided in this blog post.
This project has MIT license.