Radiogenomics analysis on Dataset from Ivy Glioblastoma Atlas Project (Ivy GAP)

What is Radiogenomics

Which part this repositories focus on?

1. GBM MRI image segmentation
2. Feature Extracion from ROI
3. Data anaylsis, including:
   • GBM subtypes prediction,which is based on Wang research on 2017, prediction (Accuracy:92%, using Multilayer Perceptron)
   • Mutual Information Analysis to Radiomics feature and gene expression energy.

File Structure

├─data # Store data, both data for training and generated results
│  ├─classification results # Classification results
│  │  ├─model
│  │  ├─Simplified
│  │  │  ├─Test
│  │  │  │  ├─Normalize
│  │  │  │  │  └─PCA
│  │  │  │  └─PCA
│  │  │  │      └─Normalize
│  │  │  └─Train
│  │  │      ├─Normalize
│  │  │      │  └─PCA
│  │  │      └─PCA
│  │  │          └─Normalize
│  │  ├─Test
│  │  │  ├─Normalize
│  │  │  │  └─PCA
│  │  │  └─PCA
│  │  │      └─Normalize
│  │  └─Train
│  │      ├─Normalize
│  │      │  └─PCA
│  │      └─PCA
│  │          └─Normalize
│  ├─gene # Gene analysis (Not important, pls ignore)
│  │  └─...
│  ├─Images # Training Images, store based on patient and MRI series
│  │  ├─W10_FLAIR_AX
│  │  └─...
│  ├─Masks # Training Masks, store based on patient and MRI series
│  │  ├─W10_AX
│  │  └─...
│  ├─Mutual_Information # Mutual Information between feature
│  ├─result # Store 
│      └─FLAIR
│          ├─monitor
│          │  ├─test
│          │  ├─train
│          │  └─validation
│          ├─ROC_curve
│          ├─test
│          └─train
│  ├─feature_extraction.csv # radiomics feature, extracted by radiomics, plz check their doc
│  ├─GBM_MRI_Dataset.csv # contianing every slice location, using for training
│  ├─gene_expression_details.csv # gene expression energy
│  ├─tumor_details.csv # Basic Database info, containing every patients' subtype
│  └─Params.yaml # using for feature extraction, configure file for pyradiomics, plz check their doc
├─exception_in_trainning # Store error message, will send to your email.
├─model # Store model
│  ├─FLAIR
│  ├─Stack
│  ├─T1
│  └─T2
├─Pictures # Pictures for README
├─requirement # Store Python environment
├─train.py  # Training Main
├─data.py # Training dataset
├─utils.py # Training utils
├─unet.py # Training net
├─trainHelper.py # Help training module
├─classification.py # Classification main
├─mutual_information.py # mutual information analysis
├─feature_extraction.py # Get radiomics features from mask
├─get_graph_csv.py # Old version file, ignore it, used to get data in csv to replot fancy graph
├─generate_gif_results.py # Get gif, old version file, only use as reference, can not run directly
└─gene_expression_monitor.py # Get some statistic information about gene expression, old version code, you can ignore it

How to use it

Segmentation

Install python environment by requirements (pip)

pip install -r requirements.txt

GBM MRI image segmentation using train.py, data.py, utils.py, unet.py, trainHelper.py. The main python script is train.py, using command to run it:

python train.py [argv1] [argv2]

argv1 is the MRI series you can choose, containing T1, T2, FLAIR, Stack. argv2 is the epoches you want.

Also, see the trainHelper.py, you can use your own mail to remind you the train's situation.

Classification

Just run classification.py

Mutual Information analysis

Just run mutual_information.py

Some results show

Segmentation

Classification

Mutual Information

TODO List

• [] Change old version generate_gif_results.py to new version