This repository contains:
-
The models for EIT image reconstruction which is published in EIT-CDAE and follow-up work.
-
Database constructed for training.
The architecture of EIT image reconstruction networks, which contains: EIT-CDAE(model A), network to denoise measurement matrix(model B) and U-net for EIT image reconstruction.
There are 2 choices of datasets: using open-source dataset: montreal dataset of EIDORS or generate simulated dataset. Both need to use EIDORS toolkit. To generate the dataset, you need run data_gen.m in dataset file. An instance of images processed is as below:
The CDAE-EIT model was built using Tensorflow, please read detail in
code/cdae.py
The network to denoise measurement matrix(model B) was built using Tensorflow. This model using measurement matrix from open-source dataset with additional noise as input, original data as output. Then the denoised data can be used to reconstruct EIT images. Please read details in
code/matricDN.py
U-net for EIT image reconstruction was built using Tensorflow, details in:
code/U-net.py
-
CDAE:
- Train and test the CDAE model with constructed database. The reconstructed images are as following:
2. Testing results of the model with open-source database: Montreal database:
- results of the matrix denoise model :
(a): Original image; (b) Image with additional noise; (c): Results of the model's output
-
U-net
1. Results of constructed database
2. Results of open-source database
- Yue Gao
- Yewangqing Lu
- Hui Li
- Boxiao Liu
- Mingyi Chen
- Guoxing Wang
- Yong Lian
- Yongfu Li*
Please cite these papers if you have used part of this work.
GAO Y, LU Y, LI H, et al. EIT-CDAE: a 2-D electrical impedance tomography image reconstruction method based on auto encoder technique[C]. IEEE Biomedical Circuits & Systems Conference, 2019: 1-4.
GAO Y, LIU B, LI H, et al. Live Demonstration: a pulmonary conditions monitor based on electrical impedance tomography measurement[C]. International Symposium on Circuits and Systems, 2019: 1-1.