DSMTA: Deeply Supervised Multi-Task Autoencoder for Biological Brain Age Estimation

🔬 About

DSMTA provides the implementation for our manuscript:

Title: Deeply Supervised Multi-Task Autoencoder for Biological Brain Age Estimation Using T1-weighted MRI

Key contributions:

• Deep supervision at multiple decoder depths for stable optimization.
• Multi-task learning, jointly training:
  1. Brain age prediction
  2. Sex classification
  3. Image reconstruction

⚙️ Features

• Robust training with auxiliary losses.
• Enhanced feature representations via multitasking.
• Ready-to-use scripts and notebooks for data prep, training, and inference.

📥 Installation

1. Clone the repository

   git clone https://github.com/PLASS-Lab/DSMTA.git
   cd DSMTA

2. Create and activate Conda environment

   conda create -n dsmta_env python=3.10 --file requirements.txt
   conda activate dsmta_env

💾 Data & Pretrained Weights

1. Pretrained Model Weights

Request access and download the pretrained weights:

https://drive.google.com/drive/folders/1k8w2HuI10wOUaSyj1qPj57HHGlVUAsVz?usp=sharing

Unzip and place all files under ./weights/.

2. OpenBHB Dataset

This study uses the OpenBHB T1-weighted MRI dataset:

• Download link: https://ieee-dataport.org/open-access/openbhb-multi-site-brain-mri-dataset-age-prediction-and-debiasing

• Mandatory citation:

  Benoit Dufumier, Antoine Grigis, Julie Victor, Corentin Ambroise, Vincent Frouin, Edouard Duchesnay, “OpenBHB: a Large‑Scale Multi‑Site Brain MRI Data‑set for Age Prediction and Debiasing”, NeuroImage, 263:119637, 2022. doi:10.1016/j.neuroimage.2022.119637

To download our processed version of OpenBHB, run:

python test_data_download.py
python train_data_download.py

This will create the directory structure:

├── data/
│   ├── train/    # Training images + labels
│   └── test/     # Testing images + labels
└── weights/      # Pretrained model files

🚀 Usage

Inference (Testing)

Launch and run the inference notebook:

jupyter notebook Inference_Notebook.ipynb

Training

Open the training notebooks in the Training_notebooks folder:

jupyter notebook Training_notebooks/

📖 Citation

If you use this code or models in your research, please cite our paper:

@article{your2025dsmta,
  title={Deeply Supervised Multi-Task Autoencoder for Biological Brain Age Estimation Using T1-weighted MRI},
  author={...},
  journal={Journal Name},
  year={2025}
}

And when using the OpenBHB dataset, also cite:

@article{dufumier2022openbhb,
  title={OpenBHB: a Large-Scale Multi-Site Brain MRI Data-set for Age Prediction and Debiasing},
  author={Dufumier, Benoit and Grigis, Antoine and Victor, Julie and Ambroise, Corentin and Frouin, Vincent and Duchesnay, Edouard},
  journal={NeuroImage},
  volume={263},
  pages={119637},
  year={2022},
  doi={10.1016/j.neuroimage.2022.119637}
}

⚖️ License

Released under the MIT License. See LICENSE for details.

---