🎉 You can now install this library locally using pip in editable mode!
💥 Check out the Installation section for more details.
Click to expand/collapse previous news.
We have reorganized the previous pipeline and opened a new python library called opensdpc. Here are some of its features:
- Corrected the usage of some functions in the previous sdpc-for-python and sdpc-linux libraries. Now they are completely aligned with the openslide library. Specifically, the following two contents have been changed:
slide.level_dimension->slide.level_dimensionsimg = Image.fromarray(slide.read_region((cut_x, cut_y), patch_level, (tile_size, tile_size)))->img = slide.read_region((x_start, y_start), args.WSI_level, (x_offset, y_offset)).convert('RGB')
- The previous method required changing the environment variables, now it is no longer neccessary to modify them manually
- The previous sdpc and openslide library have been integrated, now a unified interface is used to call them. Therefore, the opensdpc supports .sdpc and all WSI formats supported by other openslide libraries.
- Some users' machines are not compatible with the latest compiled packages. For this reason, you can still use old versions. See: old version for sdpc library.
- Due to copyright restrictions, I cannot directly provide the software for converting sdpc to svs. I am actively communicating with the company and providing an open source interface. Please contact the slide-scanner after-sales staff to request it. See sqray.com for more details.
- Provide an example for generating patches with multiple thread in order to provide pre-processing for huge datasets. See this subsection.
[Note] please use version==1.5 if you want to use sdpc-linux. I may be slow to reply, thank you for the patience 😊.
If you don't know how to deal with the old version, here is a simple plug-and-play process:
- Step1: Click the Tsinghua cloud link to download LINUX.zip directly;
- Step2: Replace the unzipped LINUX folder with the LINUX folder of the sdpc package in your own environment;
- Step3: Write the absolute path of the LINUX folder and the ffmpeg folder inside it into the environment variable, see the issue #2 for more details
- Step4: Just enter
import sdpcin python to use it.
- Updated Part "Troubleshooting"
- sdpc-linux and sdpc-win are no longer be updated. The latest library sdpc-for-python is a new version for Sdpc Python library, which can be used in both Windows and Linux Systems.
- The color correction has been updated. Now the color of the cropped patches are consistent with the color in the reading software.
- Changed function name
level_downsampletolevel_downsamples. - Chineses paths have been supported.
- Added a function to view the magnification directly:
wsi.scan_magnification(). - Added a function to view the sampling ratio directly：
wsi.sampling_rate(). - Added a function of obtaining the thumbnail image:
wsi.get_thumbnail(thumbnail_level). - The
wsi.crop_patches()function has been added. Now you can call the function directly in the code to separate the foreground tissue area and crop the patches (Using Pillow Library to save patches). - Added a option that can normalize images in
wsi.crop_patches()function, it normalizes by H and E channels.
opensdpc is a python library for processing whole slide images (WSIs) in sdpc format extended form openslide library, therefore, opensdpc also supports other common formats such as svs, tiff, ndpi, etc. To read WSIs in sdpc format in Windows platform, download the TEKSQRAY reading software.
| Download link | Extraction code | Instruction |
|---|---|---|
| Baidu Cloud | sq12 | Lite version |
| Please see the sqray.com | - | Full version |
There are several versions available. For Linux users, we now recommend using the local installation method. This is a simple way to install the library and you can see the solutions when you encounter some problems such as OSError: libDecodeHevc.so: cannot open shared object file: No such file or directory
First, clone the repo and cd into the directory:
git clone https://github.com/WonderLandxD/opensdpc.git
cd opensdpcThen, install the openslide library:
apt install openslide-tools
pip install openslide-pythonFinally, create a conda env and install the library in editable mode:
conda create -n piano python=3.10
conda activate piano
pip install -e .For Windows users, we will provide a simple way to install the library in the future.
There are two example WSIs in the data folder, which are in the sdpc format and svs format (download from CLAM) respectively.
We will use the sdpc format WSI as an example to demonstrate how to use basic functions of opensdpc.
import opensdpc
slide = opensdpc.OpenSdpc('data/20220514_145829_0.sdpc') # load the WSI
print(f'This WSI have {slide.level_count} levels.') # the number of levels
print(f'The magnification of this WSI is {slide.scan_magnification}x.') # the magnification of the WSI
print(f'The downsampling ratio of this WSI is {slide.sampling_rate}.') # the downsampling ratio of the WSI
print(f'The downsample ratio of each level is {slide.level_downsamples}.') # the downsample ratio of each level
print(f'The dimensions of each level is {slide.level_dimensions}.') # the dimensions of each level
thumbnail = slide.get_thumbnail(3) # get the thumbnail of the WSI
print(f'The thumbnail (level 3) shape is {thumbnail.shape}.') # the shape of the thumbnail
you will see the result like this:
This WSI have 4 levels.
The magnification of this WSI is 40x.
The downsampling ratio of this WSI is 0.25.
The downsample ratio of each level is (1.0, 4.0, 16.0, 64.0).
The dimensions of each level is ((26880, 21504), (6720, 5376), (1680, 1344), (420, 336)).
The thumbnail (level 3) shape is (336, 420, 3).
We provide a new easy-to-use PyTorch library with - PIANO: Pathology Image ANalysis mOdels to build a complete pipeline for histopathology image AI analysis, including generating patches, creating patch features with novel foundation models, and so on. Check out the PIANO repository for more details.
Jiawen Li, H&G Pathology AI Research Team
