Cancer Prediction

The replication of the paper Histopathology images predict multi-omics aberrations and prognoses in colorectal cancer patients.

Requirements

python==3.10
pytorch==1.12.0 
torchvision==0.13.0
tensorflow==2.10

Upon our testing on CUDA 11.6 and 11.7, the following command lines work.

conda create --name your_env_name python=3.10
conda activate your_env_name
conda install pytorch==1.12.0 torchvision==0.13.0 torchaudio==0.12.0 cudatoolkit=11.6 -c pytorch -c conda-forge
pip install tensorflow==2.10

Openslide

pip install Openslide-python

For Windows:

Download from OpenSlide, add bin and lib to environment variables.

If you encounter any errors during the import openslide process, find the file lowlevel.py from your error message, then add the code following:

import os
os.add_dll_directory("<your openslide bin path>")

For Ubuntu:

apt install python-openslide

Data Preparation

TCGA

Data source: GDC Data Portal

Filter:

• Cases
  • Primary Site: colon, rectum
  • Program: TCGA
  • Project: TCGA-COAD, TCGA-READ
• Files
  • Data Type: Slide Image
  • Experimental Strategy: Tissue Slide

Download tool: gdc-data-transfer-tool

Data Proprocessing

Refactoring

example:

python ./data_preprocessing/refactor_tcga_files.py --json_name E:/frb-cancer-prediction/TCGA20/metadata.cart.2023-11-20.json --input_path E:/frb-cancer-prediction/TCGA20 --output_path ./dataset/TCGA20/origin

Tiling

example:

python ./data_preprocessing/tiling.py --input_path ./dataset/TCGA20/origin --output_path ./dataset/TCGA20/tiled

Normalization

There are some unresolved severe problems in code, but I have not noticed any filtering procedure in paper or original code. In temperary, I selected the normalized images manually.

example:

python .\data_preprocessing\normalization.py --input_path .\dataset\TCGA20\tiled\ --output_path .\dataset\TCGA20\normalized\ --folders True

Feature Extraction

TODO

Survival Prediction

Temporary notes by wfy

I think prediction.py will run, but with my testing data it has no enough dimentions and will get error in model.

Besides, to get it run, I manually set some parameters.

In utils/prediction.py:

loss_function = weibull_loglik_discrete
optimizer = Adam(lr=0.0001)
epochs = 1

in which loss_function can be chosen from the 3 given ones in utils/prediction_model.py

In utils/prediction_data_gen.py:

n = 1
xi = 0
xj = 1

I don't know its use.

Haven't apply codes after 'evaluation'.

Code Commit Specification

It is recommended that the commit message be written in the following format:

<type>: <subject>

The message types are:

• feat - new features
• fix - fix bugs
• docs - documentations or comments
• style - code format
• refactor - refactoring, optimization (neither adding new features nor fixing bugs)
• perf - performance optimization
• test - adding tests
• chore - changes to the build process or auxiliary tools
• revert - roll back
• build - build package