A Conditional Diffusion Model for Electrical Impedance Tomography Image Reconstruction

The code in this toolbox implements the "A Conditional Diffusion Model for Electrical Impedance Tomography Image Reconstruction". More specifically, it is detailed as follow.

Training

• Simulated data: The EIDORS generated dataset is at dataset

  • Data generation: Follow the setting from Weighted SBL

• Real-world data: Two real-world data can be downloaded from UEF2017 for the 2D electrical impedance tomography dataset collected by the Finnish Inverse Problems Society at the University of Eastern Finland in 2017 (UEF2017) and KTC2023 for the Kuopio Tomography Challenge 2023.

  The real data has been placed in the ./data/ in the appropriate format.

• Put the data at ./data/ and run python main.py --mode train on 1 GPU or accelerate launch main.py --mode train on multi-GPUs machine

• Data format: The data was stored in npz format which contains ys as the voltage vector, xs as the true value of conductivity, xs_gn as the conductivity predicted by the Gauss-Newton method and TR as the conductivity predicted by the Tikhonov regression.

Test

• The pretraining weight best.pt is at pre-trained weights
• Download the pretraining weight and put it to ./results/deit/checkpoints
• Put the test data at ./data/ and run python main.py --mode test --data simulated for EIDORS generated data, python main.py --mode test --data uef2017 for UEF2017 dataset or python main.py --mode test --data ktc2023 for KTC2023 dataset
• The prediction will be at ./results/deit/checkpoints

Citation

Please kindly cite the papers if this code is useful and helpful for your research.

@ARTICLE{10922741,
  author={Shi, Shuaikai and Kang, Ruiyuan and Liatsis, Panos},
  journal={IEEE Transactions on Instrumentation and Measurement}, 
  title={A Conditional Diffusion Model for Electrical Impedance Tomography Image Reconstruction}, 
  year={2025},
  volume={},
  number={},
  pages={1-1},
  keywords={Conductivity;Image reconstruction;Electrical impedance tomography;Diffusion models;Noise reduction;Accuracy;Training;Inverse problems;Feature extraction;Deep learning;Electrical impedance tomography;image reconstruction;diffusion model;probabilistic model;measurement visualization},
  doi={10.1109/TIM.2025.3550245}}

Acknowledgement

This code is mainly built upon DiT repositories.

Contact Information:

If you encounter any bugs while using this code, please do not hesitate to contact us.

Shuaikai Shi [[email protected]]