RIED-Net

Official PyTorch implementation of Deep Residual Inception Encoder-Decoder Network for Amyloid PET Harmonization [Alzheimer's & Dementia]

Jay Shah^1,2, Fei Gao^1,2, Baoxin Li^1,2, Valentina Ghisays³, Ji Luo³, Yinghua Chen³, Wendy Lee³, Yuxiang Zhou⁴, Tammie L.S. Benzinger⁵, Eric M. Reiman³, Kewei Chen³, Yi Su^1,2,3, Teresa Wu^1,2

¹ASU-Mayo Center for Innovative Imaging, ²Arizona State University, ³Banner Alzheimer’s Institute, ⁴Dept of Radiology, Mayo Clinic, Arizona, ⁵Mallinckrodt Inst. of Radiology, Washington University

---

Multiple positron emission tomography (PET) tracers are available for amyloid imaging, posing a significant challenge to consensus interpretation and quantitative analysis in Alzheimer's disease research. We accordingly developed and validated a deep learning model as a harmonization strategy. Learn more about its practical applications in ASU's blog post: Using AI to battle Alzheimer's.

Visual comparison of synthetic images generated using RIED-Net to real Pittsburgh Compound-B (PIB) data for the OASIS (A & B) and GAAIN (C & D) datasets

Installation

Instructions to install MONAI can be found here and the appropriate version of Pytorch using locally. Packages used in the current version of this code.

monai==1.3.0
torch==2.1.0+cu118
torchaudio==2.1.0+cu118
torchmetrics==1.2.0
torchvision==0.16.0+cu118

Dataset

You should structure your aligned dataset in the following way:

trainfold/
  ├── train
    ├──images
      ├──xxx.nii
      ├──...
    ├──targets
      ├──yyy.nii
      ├──...
  ├── val
    ├──images
      ├──xxx.nii
      ├──...
    ├──targets
      ├──yyy.nii
      ├──...

Training

python train.py --dataset trainfold1 --batch_size 3 --model_name resunet

Evaluation

testfold should have same structure as trainfold. Checkout prepare_data.py to prepare the data folds.

python generate.py --dataset testfold1 --model_name resunet

Citation

Please consider citing RIED-Net if this repository is useful for your work.

@article{shah2022deep,
  title={Deep residual inception encoder-decoder network for amyloid PET harmonization},
  author={Shah, Jay and Gao, Fei and Li, Baoxin and Ghisays, Valentina and Luo, Ji and Chen, Yinghua and Lee, Wendy and Zhou, Yuxiang and Benzinger, Tammie LS and Reiman, Eric M and others},
  journal={Alzheimer's \& Dementia},
  volume={18},
  number={12},
  pages={2448--2457},
  year={2022},
  publisher={Wiley Online Library}
}

Acknowledgments

This research has been supported partially by NIH grants R01AG031581, R01AG069453, P30AG019610, and Arizona Department of Health Services (ADHS) and the State of Arizona, ADHS Grant No. CTR040636. This is a patent-pending technology.