The dataset is composed of Hematoxylin and eosin (H&E) stained breast histology microscopy and whole-slide images. Challenge participants should evaluate the performance of their method on either/both sets of images.
1. Microscopy images
Microscopy images are labelled as normal, benign, in situ carcinoma or invasive carcinoma according to the predominant cancer type in each image. The annotation was performed by two medical experts and images where there was disagreement were discarded.
|Normal||Benign||in situ carcinoma||Invasive carcinoma|
The dataset contains a total of 400 microscopy images, distributed as follows:
- Normal: 100
- Benign: 100
- in situ carcinoma: 100
- Invasive carcinoma: 100
Microscopy images are on .tiff format and have the following specifications:
- Color model: R(ed)G(reen)B(lue)
- Size: 2048 x 1536 pixels
- Pixel scale: 0.42 µm x 0.42 µm
- Memory space: 10-20 MB (approx.)
- Type of label: image-wise
Note that the microscopy image dataset is an extension of the one used in this article, made publicly avaiable at this repository. For the BACH challenge, please download the new, more complete dataset, via the DOWNLOAD tab.
The patient-wise origin of each microscopy image is partially available on the following file. The anonimization process does not allow to retrieve the origin of all images, but it is safe to assume that all undentified images come from different patients of the identified ones.
2. Whole-slide images
Whole-slide images are high resolution images containing the entire sampled tissue. In this sense, microscopy images are just details of the whole-slide images. Because of that, each whole-slide image can have multiple normal, benign, in situ carcinoma and invasive carcinoma regions. The annotation of the whole-slide images was performed by two medical experts and images where there was disagreement were discarded. Each image has a corresponding list of labelled coordinates that enclose benign, in situ carcinoma and invasive carcinoma regions (the remaning tissue is considered normal and thus is not relevant for performance evaluation).
Whole-slide images are on .svs format and have the following specifications:
- Color model: R(ed)G(reen)B(lue)
- Size: variable (eg: 42113 x 62625 pixels)
- Pixel scale: 0,467 µ/pixel
- Memory space: 8 GB (approx.) when in numpy array (pyhton), 200-250 MB (approx.) in .svs
- Acquisition system: Leica SCN400
- Type of label: pixel-wise
Annotations of the coordinates of the points that enclose each region are made avaliable as a .xml file.
The patient-wise origin of the WSI is provided in the following file.
Reading the dataset
Microscopy images are available in .tiff format.
Whole-slide images are available in .svs format.
For convinience, a Python script for reading the .svs and the .xml containing for the annotations of the whole-slide images is available together with the remaining data.
The provided scripts require OpenSlide.
For Windows users:
- open a command-line and install openslide-python: pip install openslide-python
- download the Windows binaries here;
- unzip the binaries and add the bin folder to your path variable.
For Linux users:
- open a terminal and install openslide-python: pip install openslide-python
- install openslide: sudo apt-get install python-openslide
Requires opencv version 3.2.0 (at least);
The .xml reading script returns a list containing, for each region of interest:
- the coordinates of the perimeter of the region (X,Y -- columns,rows in Python indexing - origin at the left upper corner)
- the label of the region
- the perimeter of the region, in micrometers
- the area of the region, in square micrometers
- the pixel scale, micrometers/pixel