This is a public repository for "Fully Automated Hybrid Network to Predict IDH Mutation Status of Glioma via Deep Learning and Radiomics" by Choi et al. The core code is entirely based on yoonchoi-neuro/automated_hybrid_IDH.
The automated hybrid model consists of UNet-based Model1 for tumor segmentation, ResNet-based Model 2 for IDH status prediction, and automated processing pipeline inbetween. Model 2 integrates 2D MR images, radiomic features of 3D tumor shape & loci, and age in one CNN.
I've modified and adapted the code into a script in order to test multiple patients on a single run. I've added the conversion part from DICOM .dcm to NIFTI nii.gz using the dcm2niix tool by rordenlab. I've also added the skulstripping part that is missing from the original repo.
- Python module requirements : Nipype / FSL / ANTs / PyRadiomics / PyTorch
- The process resquires GPU.
- To test your cases, create inside
./INPUTa separate directory for each patient and rename it to its uniqueid. Inside each patient's directory put the 3 axial MRI DICOM directories renamed to:T1C,T2andFLAIR. - Edit the
age.csvfile and populate it with your patients'ageandid. - Run
main.py - The script outputs a
predict.csvfile inside./OUTPUTwhere all prediction scores are listed alongside patient'sid.
-
Inside
Splitdirectory the script was split into two to separate image proccesing and MODEL1 from MODEL2. -
The code to test one sample using Jupyter-Notebook is avaialble at https://github.com/ihrapsa/automated_hybrid_IDH
yoonchoi-neuro - for the hard work she put into this
teo2mirce - for all the support
💬 discord: jonah1024#4422
📧 email: [email protected]