A project to create an easy-to-use way of finding areas of common patterns and structures within an image. Should work on any image, designed for use on any biological images including DAPI, IMC and H&E.
Ben's attempt to make a parallel version using joblib is in the other branch 'B_parallel'. The B_parallel version is faster and uses less disk space compared with the main branch, so is recommended.
A test image file is current located here: https://zenodo.org/record/7821268#.ZDaHS3bMIuU (this would require changes to the code, or you can use the image provided in private google drive)
At the moment the pipeline is set up to run without a groundtruth included by commenting out the relevant sections.
To install the environment, download the environment_lbp3d.yml file or the environment_lbp3b_windows.yml file
then run
conda env create -f environment_lbp3d.yml or on windows conda env create -f environment_lbp3b_windows.yml
then
conda activate lbp3d
or
conda activate lbp3b_windows
Step 1: Run 2021-07-06_Local_binary_pattern_on_images.ipynb
Use this script to run Local Binary Pattern over an image. This outputs individual files for each condition where the x, y coordinates are for each patch and the intensity contains the LBP texture value. Multiple images are created to hold the LBP information from different radii.
Step 2: Run 2022-10-28b_Creating_patch_LBP_signatures_with_channels_from_LBP_values_that_have_same_dim_as_img.ipynb
Combines the different LBP conditions into one file for easier analysis.
Step 3: Run 2022-11-01_UMAP_each_image_separately.ipynb or 2022-11-10_UMAP_and_clustering.ipynb
Either of these can be run independently of which options you chose above
Run clustering and analysis in an unsupervized way.
Step 4 (optional): If you ran 2022-11-10_UMAP_and_clustering.ipynb then you will have output some clusters that can be displayed on an image that can be viewed in napari with this script: 2022-11-10d_View_cluster_masks_in_napari.ipynb