Deformation-Based Morphometry using Deep Learning

This repository contains the source code for the research paper titled "A deformation-based morphometry framework for disentangling Alzheimer's disease from normal aging using learned normal aging templates". You can find the paper here.

TODOs

• Add the link to share the OASIS-3 Dataset
• Add the link to the paper
• Share the pre-trained weights (in ./models)
• Add the Jupyter notebook demo (step-by-step_example.ipynb)

OASIS-3 Dataset (TBA)

We provide preprocessed OASIS-3 data as part of this study. The data is included in the npz format, along with corresponding segmentation masks generated using SynthSeg version 2.0.

If you use this dataset, please cite the following and refer to the corresponding Data Use Agreement .

• OASIS-3: Longitudinal Neuroimaging, Clinical, and Cognitive Dataset for Normal Aging and Alzheimer Disease. Pamela J LaMontagne, Tammie L.S. Benzinger, John C. Morris, Sarah Keefe, Russ Hornbeck, Chengjie Xiong, Elizabeth Grant, Jason Hassenstab, Krista Moulder, Andrei Vlassenko, Marcus E. Raichle, Carlos Cruchaga, Daniel Marcus, 2019. medRxiv. doi: 10.1101/2019.12.13.19014902

Pre-trained Healthy Template Creation Models

We offer pretrained Atlas-GAN weights trained on the OASIS-3 Dataset for simulating normal aging. Additionally, we provide inference scripts for extracting the learned diffeomorphic registration module and template generation module. (in ./models)

The extracted healthy templates are also available in NIfTI format, spanning ages from 60 to 90. (in ./example/CN_templates)

Citation

@misc{fu2023deformationbased,
      title={A deformation-based morphometry framework for disentangling Alzheimer's disease from normal aging using learned normal aging templates}, 
      author={Jingru Fu and Daniel Ferreira and Örjan Smedby and Rodrigo Moreno},
      year={2023},
      eprint={2311.08176},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Acknowledgements:

This repository is developed based on the Atlas-GAN project and makes extensive use of the VoxelMorph library.