A tool for the hypothalamus and its subregions capable of working across different MRI modalities and resolution. H-SynEx was trained on synthetic images derived from high-resolution ex vivo MRI.
The model was trained on synthetic images derived from ex vivo MRI label maps:
Example of testing images:
Labels
| Label | Subregion |
|---|---|
| 1 | left anterior inferior |
| 2 | left posterior |
| 3 | left tuberal inferior |
| 4 | left tuberal superior |
| 6 | left anterior superior |
| 7 | right anterior inferior |
| 8 | right posterior |
| 9 | right tuberal inferior |
| 10 | right tuberal superior |
| 12 | right anterior superior |
To use the tool, please follow the following steps:
-
Download this repository
-
Using the terminal, enter the inference folder:
cd <your_path>/hsynex/inference -
Run the 'find_hypothalamus_subnuclei.py':
python --input_path <input_path> --out_path: <out_path>
Where:
<input_path>is the path where the input MR images are located<out_path>is the chosen path to save the segmentations
This code was implemented using Python 3.10. To run it, install the dependencies listed at requirements.txt.
This repository also relies on a modified implementation of pytorch-3dunet, which is already included.
Finally, it is necessary to install FreeSurfer 7.4
If you use H-SynEx in your project, please cite:
[arXiv]: Rodrigues, Livia, et al. "H-SynEx: Using synthetic images and ultra-high resolution ex vivo MRI for hypothalamus subregion segmentation." arXiv preprint arXiv:2401.17104 (2024).