py-imcyto

IMC segmentation in python.

pyimcyto is a python library for nuclear and cellular segmentation in imaging mass cytometry images.

The segmentation model uses a U-net++ archiecture with autoencoder-based anomaly detection to refine predictions, and is the same as the deep-imcyto segmentation model which is distributed as part of TRACERx-PHLEX -- a Nextflow-based multiplexed image analysis pipeline, however the implementation is designed for those who wish to use the model itself directly in their own python code.

Running in Google colab

The easiest way to get started with pyimcyto is to use the py-imcyto Google Colab notebook. Click below to run in Google Colab and learn to segment nuclei and cells in IMC images with py-imcyto.

• Simply follow the steps below to do basic prediction with the trained deep-imcyto model.
• Use a GPU runtime in Google Colab to facilitate faster prediction.

---

Getting started

Alternatively, perform the following steps to install py-imcyto in your own Python environment.

1. Download the trained deep-imcyto model weights from our Zenodo repository

2. Clone the pyimcyto repository

3. Create a new environment and install the necessary dependencies:

conda create -n deepimcyto python=3.8 numpy pandas scipy matplotlib scikit-image cudatoolkit=11.8.0 scikit-learn -c conda-forge -c anaconda

Install Tensorflow and other dependencies from pip:

pip install nvidia-cudnn-cu11==8.6.0.163 tensorflow==2.12.* jupyter opencv-python cupy-cuda11x tqdm

Usage

from pyimcyto.seg import deepimcyto
import skimage.io as io

# define the path to the deep-imcyto model weights directory:
weights = './weights'

# create the deep-imcyto model with the trained weights:
dicy = deepimcyto(weights)

# read in an IMC image:
img_path = './data/P1_TMA006_L_20190619-roi_6.ome-roi_p1_tma006_l_20190619-roi_6.png'
img = io.imread(img_path)

# predict nuclei:
result = dicy.predict(img)

# visualize results:
dicy.get_boundaries() # get boundaries of instance mask
dicy.make_overlays() # overlay on test image

# Plot result:
fig, ax = plt.subplots(1, 1, figsize=(10, 10))
ax.imshow(overlay[overlay.shape[0]//2 - 256:overlay.shape[0]//2 + 256, overlay.shape[1]//2 - 256:overlay.shape[1]//2 + 256]) # plot 512x512 tile in centre of test image
ax.set_axis_off()
plt.tight_layout()

See the example jupyter notebook for more examples of usage.

Citation

Credits

deep-imcyto/py-imyto is primarily developed by Alastair Magness at The Francis Crick Institute.