Electrical impedance tomography (EIT) with neural networks.
This repo contains the source code to reproduce the results from those papers
A Post-Processing Method for Three-Dimensional Electrical Impedance Tomography, Martin, S., Choi, C.T.M. A Post-Processing Method for Three-Dimensional Electrical Impedance Tomography. Sci Rep 7, 7212 (2017) https://doi.org/10.1038/s41598-017-07727-2
A novel post-processing scheme for two-dimensional electrical impedance tomography based on artificial neural networks, Martin, S., Choi, C.T.M. A novel post-processing scheme for two-dimensional electrical impedance tomography based on artificial neural networks. PLOS ONE 12(12): e0188993 https://doi.org/10.1371/journal.pone.0188993
A Feasibility Study on Real-Time High Resolution Imaging of the Brain Using Electrical Impedance Tomography, Martin, S., A Feasibility Study on Real-Time High Resolution Imaging of the Brain Using Electrical Impedance Tomography, ArXiv 2202.13159 https://arxiv.org/abs/2202.13159
A Post-Processing Tool and Feasibility Study for Three-Dimensional Imaging with Electrical Impedance Tomography During Deep Brain Stimulation Surgery, Martin, S., A Post-Processing Tool and Feasibility Study for Three-Dimensional Imaging with Electrical Impedance Tomography During Deep Brain Stimulation Surgery, ArXiv 2204.05201 https://arxiv.org/abs/2204.05201
Download EIDORS from the original SVN repo: http://eidors3d.sourceforge.net/
Download MatGeom from Github (included as a submodule): https://github.com/mattools/matGeom
Install Netgen: https://ngsolve.org/downloads
Add EIDORS and MatGeom into your Matlab path, add this repository into your Matlab path as well
The demo directory contains the entry points to reproduce the results. You can update the parameters at the beginning of the file
You can choose whether to add noise, distortion, the target type, them the script train the models, generate some additional random targets and displays the results
Similar to 2D EIT. Note that you may need a huge amount of memory to solve some large problems. Finite elements models with a smaller number of nodes have been designed for rapid testing, but those models won't give any good result due to coarsity of the models.
This code was originally developed for different versions of Matlab, from R2013a from R2016a, and successfully tested again under version R2020b
Required: Neural networks toolbox
Parallel computing toolbox can make things faster, although it was crashing on latest Matlab version Communications toolbox and Symbolic math toolbox are required to use the noise functions
EIDORS versions 3.8 to 3.10 are working correctly
Version 1.2.2
Tested with Netgen 6.2 https://ngsolve.org/downloads