SCSR-core

Official implementation of Stochastic Cortical Self-Reconstruction (SCSR).

Installation

1. Check out repository
2. Create environment: conda env create --file environment.yml
3. Activate environment: conda activate SCSR
4. Download model https://drive.google.com/file/d/1qmD5m3wR1F_sqVmBTyZgWWxi_VCpEbKz/view?usp=sharing and copy to directory checkpoints

Data

We used data from Alzheimer's Disease Neuroimaging Initiative (ADNI).

Usage

• The package uses PyTorch
• As input data, an input table with columns ['DX', 'AGE', 'PTGENDER', per-vertex values] is expected as a .feather file
• To train SCSR, set the path to the input table table_path in the training file and call python SCSR_train.py config_files/training_configs/config.yaml.
• For sampling, again set the path to the input table table_path in SCSR_sample.py. We provide a minimal working example by default using data/sample_data.feather, which can be run by calling python SCSR_sample.py. The resulting cortex maps & Z-scores can be visualized on FreeSurfer surfaces, e.g., using visgeom.

Citation

@article{wachinger2025individualized,
  title={Individualized mapping of aberrant cortical thickness via stochastic cortical self-reconstruction},
  author={Wachinger, Christian and Hedderich, Dennis M and Thalhammer, Melissa and Bongratz, Fabian},
  journal={Medical Image Analysis},
  pages={103788},
  year={2025},
  publisher={Elsevier}
}