TensorFlow cunstom runner to read micro images from whole slide images.
This package uses openslide-python(https://openslide.org/api/python/)'s read_region function to produce tile images from whole slide images.
By using this library, you can input small patch images to your TensorFlow model. For example, let's suppose you have trained your machine learning model to detect mitotic cells of breast cancer and want to apply the trained model to the entire region of new pathology slides, the following codes would be the solution.
import tensorflow as tf
from openslidertfpy import MicroPatchReader
openslide_read_region_params = [
((100, 100), 0), # location and level for openslide's read_region func
((200, 200), 0),
((300, 300), 1) # And so on...
]
image_width, image_height = 128, 128
with tf.Graph().as_default():
coordinator = tf.train.Coordinator()
reader = MicroPatchReader(
"sample.svs", coordinator, image_width, image_height
)
images, locations_batch, levels_batch = reader.get_inputs()
# Define images, openslide's read_region function parameters
results = some_op(images) # Place your ML model here
with tf.Session() as sess:
# Some function to initialize values should be placed here
# Start reading pathology images
runner.start_thread(openslide_read_region_params)
while not coordinator.should_stop():
actual_results, actual_locations, actual_levels = sess.run(
[results, locations_batch, levels_batch]
)
See examples/read_region.py.
Please install openslide and tensorflow to your system first.
On Ubuntu;
$ sudo apt install python3-openslide
$ pip install tensorflow # or tensorflow-gpu$ python3 -m pip install git+https://github.com/OtaYuji/openslider-tfpyCheck the installation.
$ python3 -c "import openslidertfpy; print(openslidertfpy.__version__)"or you can simple place openslidertfpy/openslidertfpy.py file to your working directory.