Integrating Model-based Reconstruction and Deep Learning for Accelerating Mass Spectrometry Imaging

Official implementation of "Integrating Model-based Reconstruction and Deep Learning for Accelerating Mass Spectrometry Imaging".

Citation

If you find our work or this repository useful, please consider giving a star ⭐.

@article{PnPMSI2025,
  title={Integrating Model-based Reconstruction and Deep Learning for Accelerating Mass Spectrometry Imaging},
  author={Mithunjha Anandakumar and Timothy Trinklein and Stanislav S. Rubakhin and Jonathan Sweedler and Fan Lam},
  journal={Analytical Chemistry},
  volume={97},
  issue={45},
  pages={25120-25128},
  year={2025},
  doi = {10.1021/acs.analchem.5c04075}
}

This repository contains the implementation of the UNet-based denoiser model as a regularizer and its integration with model-based reconstruction under the plug-and-play framework.

Dataset

Find the dataset used in our work at : https://databank.illinois.edu

Getting Started

Installation

The algorithms were developed in the Pytorch Environment : https://pytorch.org/.

pip install torch==1.10.0+cu113 torchvision==0.11.1+cu113 torchaudio===0.10.0+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html

Run the code below to install all other dependencies.

pip install -r requirements.txt

Training of the Denoiser

Use the following code to train the denoiser model.

python3 training.py --experiment_name <EXPERIMENT_NAME> --epochs <#EPOCHS> --data_directory <PATH_TO_DATA> --save_path <PATH_TO_A_FOLDER> --noise_level_range <[LOWER, UPPER]> 

Plug and Play for sparse sampled ion images

Use the following code to reconstruct the sparse sampled data for any sampling percentage.

python3 run.py --sampling_percentage <SAMPLING_PERCENTAGE> --N <IMAGE_DIMENSION> --max_iteration <MAX_ITER> --model_path <MODEL_PATH> --save_path <OUTPUT_PATH>

Sample output

The quantitative and qualitative output for the retrospectively sampled mouse brain FT-ICR data dataset case follows. Evaluation metric (50% sampling): mean value for the entire test dataset (standard deviation)

SSIM : 0.8079 (0.02536)

PSNR : 28.05457 (1.93238)