README.md

Performing a 2D Nucleus Segmentation With Cellpose and StarDist on Xenium Data

Overview

This document provides a detailed overview of the processes and resources utilized for performing 2D nucleus segmentation using Cellpose and StarDist on Xenium data.

Stardist:

Download weights and resources in a .zip file, decompress the folder and add the project folder inside the repo (same level as all the other folders eg;inputs, masks): https://drive.google.com/file/d/1CpCZ1GZamGl3kcfPDySYkWi4fgI56g3O/view?usp=sharing

Files Description

cross_validation_stardist

This script is used for cross-validation of the StarDist model. It contains routines to systematically evaluate the model's performance across different subsets of the data.

visualisation_stardist

This script is dedicated to visualizing the data in the context of the StarDist model. It provides functions to plot the data, model predictions, and potentially compare these with ground truth annotations.

stardist_model

This script includes the code for training the StarDist model. It sets up the model architecture, compiles it with the necessary configurations, and trains it on the provided dataset.

stardist.ipynb

THIS SHOULD BE THE FIRST FILE YOU RUN TO DOWNLOAD THE DEPENDENCIES

This Jupyter notebook serves as a comprehensive guide and demonstration of the entire workflow. It walks through cross-validation, data visualization, model training, and evaluation, providing a hands-on experience with the StarDist model. You can choose what you want to try by changing the global variable at the beginning of the notebook

Cellpose: Resources and Files

Below is a comprehensive list of the key folders and files integral to the Cellpose training and evaluation:

Notebooks

• CellposeTrainingArena.ipynb: Contains code for training models and collecting data. It's pivotal for cross-validation across various learning rates and epochs.
• cellpose.ipynb: Dedicated to the evaluation of our trained Cellpose model, this notebook includes several visualizations.

Training Directory: cellposetraining/

This directory encompasses all the necessary data for training and validating the Cellpose model.

Subdirectories and Content

• cellposetraining/control/:

  • Contains images and masks for validating the trained Cellpose model.
  • final_segmentation/: Includes the Cellpose-produced masks for mip_2.tif.
  • images/: Contains the mip_2.tif image.
  • masks/: Houses manually segmented/ground truth masks for mip_2.tif.

• Validation Data:

  • cellposetraining/{morphology}_validate_mask and cellposetraining/{morphology}_validate: These folders contain the validation data used during Cellpose training.

• Human-in-the-Loop Training Data:

  • cellposetraining/traindataHIL and cellposetraining/trainingdataHIL_mip: Includes training data generated from our Human-in-the-Loop (HIL) iterations.