Enhancing Brain Age Estimation with a Transformer Add-on Incorporating a Longitudinal Consistency Prior

Getting Started

1. Data Preparation

• Prepare your CSV files with paths to input feature maps (extracted with pretrained encoders) and metadata (see example in data/dataset.csv).
• Update paths and parameters in config.py and config_inference.py as needed.
• Update paths in bias_correction/coles_bias_correction.py

2. Training

Run the main training script:

python train.py

• Supports both cross-sectional and longitudinal training (set in config).

3. Inference

Run inference on validation, test, or external datasets:

python inference/inference_main.py

• Results and metrics are saved to the experiment folder.
• Plots of predicted vs. real ages are generated.

4. Bias Correction

Apply post-hoc bias correction to predicted ages:

python bias_correction/coles_bias_correction.py

• Uses Huber regression to correct for age bias (Coles et al.).
• Saves corrected predictions and updated metrics.

Key Modules

• model_module.py: Defines model architectures and training logic.
• data_module.py: Handles data loading, batching, and custom sampling for longitudinal pairs.
• utils.py: Preprocessing, normalization, augmentation, and helper functions.
• train.py: Orchestrates training, logging, and checkpointing.
• inference/inference_main.py: Loads trained models and runs inference.
• bias_correction/coles_bias_correction.py: Applies bias correction to predictions.

Citation

If you use this codebase in your research, please cite the article 'Enhancing Brain Age Estimation with a Transformer Add-on Incorporating a Longitudinal Consistency Prior'.

License

This project is released under the MIT License.

---

Contact:
For questions or contributions, please open an issue or contact me at clara.lisazo@udg.edu.