uw-loci / collagen-fiber-metrics

Deep generative models for collagen fiber centerline segmentation and extraction in cancer tissue

Collagen fiber extraction and analysis in cancer tissue microenvironment

This repository is part of the paper

Collagen Fiber Centerline Tracking in Fibrotic Tissue via Deep Neural Networks with Variational Autoencoder-based Synthetic Training Data Generation,
Hyojoon Park*, Bin Li*, Yuming Liu, Michael S. Nelson, Helen M. Wilson, Eftychios Sifakis, Kevin W. Eliceiri,
Medical Image Analysis 2023.

Related repositories:

• DuoVAE

• Collagen fiber synthesis and extraction networks

Installation

Install anaconda/miniconda

  $ conda env create --name collagen --file env.yml
  $ conda activate collagen

If there are issues with OpenCV

  $ pip install opencv-contrib-python

Install ridge-detection package

  $ pip install ridge-detection

Install PyTorch

CenterLine class

This class handles the conversion between a centerline mask and a dictionary that contains the coordinates of individual centerlines, as well as the fiber property computation.
For example, create a CenterLine object using a binary centerline mask:

    from centerline import CenterLine
    centerline = CenterLine(centerline_image=io.imread("examples/example_centerline.png"), associate_image=io.imread("examples/example_image.tif"))

centerline_image is a binary centerline mask, associate_image is an optional image of collagen fiber to which the binary mask corresponds.
Compute the fiber properties:

    centerline.compute_fiber_feats() 
    print(dict(list(centerline.feats.items())[:-1]))

Create a colorized overlay of fiber centerline instances on the collagen fiber image:

    centerline_res.create_overlay()

Other ways to create a CenterLine object, check notebook centerline-basics.ipynb.

Read CT-FIRE results or use ridge detection

Check notebook centerline-baselines.ipynb

FiberExtractor class

This class handles the computation of a fiber centerline mask from a neural network.
Process a collagen fiber image:

    from fiber_extraction import FiberExtractor, UNet
    from skimage import io, img_as_uint
    net = UNet(1, 1, 16, True).eval()
    net.load_state_dict(torch.load('weights/netG.pt'))
    fiber_extractor = FiberExtractor(net)
    im_arr = img_as_uint(io.imread('examples/test_input.png'))
    result = fiber_extractor.compute(im_arr)

Compute the normalization range for 16-bit images. file_list is a list of directories to the image files.

    fiber_extractor.normalization_range(file_list=file_list)

This function computes the range in the 16-bit image set to be stretched to the range of (0, 65535).

Citation

@article{park2023collagen,
         title={Collagen fiber centerline tracking in fibrotic tissue via deep neural networks with variational autoencoder-based synthetic training data generation},
         author={Park, Hyojoon and Li, Bin and Liu, Yuming and Nelson, Michael S and Wilson, Helen M and Sifakis, Eftychios and Eliceiri, Kevin W},
         journal={Medical Image Analysis},
         volume={90},
         pages={102961},
         year={2023},
         publisher={Elsevier}
}