Scripts and model accompanying our paper "Improving Automated Glioma Segmentation in Routine Clinical Use Through Artificial Intelligence-Based Replacement of Missing Sequences With Synthetic Magnetic Resonance Imaging Scans" (Invest Radiol. 2022 Mar 1;57(3):187-193.).
The program can synthesize missing MR modalities to enable downstream processing of glioma examinations.
generator.h5: The fully trained model used for evaluation in our manuscript.
create_syn_image.py: This script will create missing MR sequences (T1, T1ce, T2 or FLAIR) from available input modalities. This script accepts as input parameters the file paths to all four "BraTS" sequences:
-t2 /path/to/t2.nii.gz
-t1 /path/to/t1.nii.gz
-t1c /path/to/t1c.nii.gz
-flair /path/to/flair.nii.gz
All missing modalities are synthesized and stored in the path given above. Please ensure that all input images are correctly preprocessed (like in the example; you can use BraTS.Toolkit for that). The file "generator.h5" must be in the same directory as the script. To for example create a synthetic flair from the example (assuming the files are stored in data/) call:
python3 create_syn_image.py -t1 data/example_t1.nii.gz -t1c data/example_t1c.nii.gz -t2 data/example_t2.nii.gz -flair data/example_flair_synth.nii.gz
/data/: Four example NIfTIs, already fully preprocessed with BraTS.Toolkit.
/training/mmgan_brats.py: Code used for training the GAN.
The program works best with co-registered and skullstripped files in SRI-24 space. These can be obtained from BrainLes or BraTS Toolkit
If you use this program, please cite to support our development:
Thomas, M. F., Kofler, F., Grundl, L., Finck, T., Li, H., Zimmer, C., Menze, B., & Wiestler, B. (2022). Improving Automated Glioma Segmentation in Routine Clinical Use Through Artificial Intelligence-Based Replacement of Missing Sequences With Synthetic Magnetic Resonance Imaging Scans. Investigative radiology, 57(3), 187–193. https://doi.org/10.1097/RLI.0000000000000828
@article{thomas2022improving,
title={Improving automated glioma segmentation in routine clinical use through artificial intelligence-based replacement of missing sequences with synthetic magnetic resonance imaging scans},
author={Thomas, Marie Franziska and Kofler, Florian and Grundl, Lioba and Finck, Tom and Li, Hongwei and Zimmer, Claus and Menze, Bj{\"o}rn and Wiestler, Benedikt},
journal={Investigative Radiology},
volume={57},
number={3},
pages={187--193},
year={2022},
publisher={LWW}
}