LabelSlide is a slide annotation tool and label object bounding boxes in virtual slides (generally used in pathology)
LabelSlide is based on B/S structure and written in Python.
Using OpenSlide to read the virtual slides and create DeepZoom Image.
Then view the zoomable image in browser by OpenSeadragon.
Adding Fabric.js canvas to draw label box overlaying the OpenSeadragon viewer.
The webpage build with Bootstrap
Python
pip install -r requirements.txt
python label_slide.py
access 127.0.0.1:5000 (as default)
If show error "Could not find module 'libopenslide-0.dll'". Please download libvips and unzip to bin folder
There's two config file.
app_config.py
Including HOST, PORT for the app and the SLIDE_CONFIG for slide config(the file path of config.yaml).
ps: parameter name must be uppercase
config.yaml
Containing the following parameters
slide_dir: the directory where slides are kept.
label_dir: the directory where the annotation will be saved to.
class: pre-defined classes including id, name and color.
For example:
slide_dir: D:\Dataset\slides
label_dir: D:\Dataset\labels
class:
- id: 0
name: A
color: '#2196f3'
- id: 1
name: B
color: '#0014a8'
ps: the class id must be integer and the color must be hex format.
All content in config.yaml could be modified from webpage
In theory, the tool can read virtual slides in several formats because using OpenSlide:
- Aperio (.svs, .tif)
- Hamamatsu (.vms, .vmu, .ndpi)
- Leica (.scn)
- MIRAX (.mrxs)
- Philips (.tiff)
- Sakura (.svslide)
- Trestle (.tif)
- Ventana (.bif, .tif)
- Generic tiled TIFF (.tif)
But only try MIRAX(.mrxs) under development.
If you want to read slide in other formats, you should modify code in label_tool.py by yourself.
Support rectangle box and polygon box currently.
Hotkeys
| Key | Function |
|---|---|
| q | View Mode |
| e | Create new rectangle box |
| r | Create new polygon box |
| del | Delete the selected rect box |
Annotation file
One slide corresponds to one annotation file.
Take annotation for MIRAX (.mrxs) as an example
slide: C20200174
boxes:
- class: 0
type:rect
x: 30749
y: 70121
w: 468
h: 434
- class: 1
type: poly
points:
- x: 33813
y: 62176
- x: 33931
y: 62571
- x: 34099
y: 62308
x,y is the top left pixel and w,h is the region size in the level 0 reference frame(highest resolution)