deep-stroma-histology's Introduction

deep-stroma-histology

Convolutional neural networks for extracting a "deep stroma score" from histological images of human cancer

What is this?

This is Matlab code to train a convolutional neural network for tissue classification in histological images of human cancer. This network can be used to derive a "deep stroma" risk score from such images. Also, this repository contains R code that we used for downstream statistics. The methods are described in our paper "A deep learning based stroma score is an independent prognostic factor in colorectal cancer"

What do I need to get started?

You need the code (provided in this repository) and the images which are available for download here: http://doi.org/10.5281/zenodo.1214456 We used the normalized 100K data set for training, but you can also download the non-normalized 100K data set.

Also, you need to install the "color normalization toolbox" from this link: https://warwick.ac.uk/fac/sci/dcs/research/tia/software/sntoolbox/ You should install it in the sub-folder "subroutines_normalization"

Where can I get your pre-trained VGG model?

The model is available here: http://doi.org/10.5281/zenodo.1420524

deep-stroma-histology's People

Stargazers

Watchers

deep-stroma-histology's Issues

why set it this value?

reproduce problem

Thank you for your hard work and code sharing. But I found that I cannot reproduce your result in Fig 5. After running your SCORE_APPLY_v1.R script, I got CAF score, Stage I-IV equal 0.772, which is not equal to your result in Fig 5. please help!

"color normalization toolbox" unavailable

The link to the "color normalization toolbox" (https://warwick.ac.uk/fac/sci/dcs/research/tia/software/sntoolbox/) is currently redirecting to the login page, making it inaccessible. I was able to find following alternative link (https://warwick.ac.uk/fac/cross_fac/tia/software/sntoolbox/), but I am not sure if this is the used implementation.

Could you kindly share the original toolbox used within the repository to facilitate proper utilization and reproducibility? Thank you.

stroma percent value mismatch

for example, TCGA-CM-6675-01A-01-TS1.2bea1d2f-2ed2-4179-a3e3-2ea1c135d638.svs, the percent_stromal_cells in your paper is 12%, but in TCGA GDC I found the value 10%. I wonder if it's because the update of TCGA website? looking forward to your reply.

mask9toRGB function not found

I was trying to run the code using the trained model and while running the file deploy_deep_texture_classifier, I am getting the error that mask9toRGB function is not found. Can you please help to resolve this issue?