※ This project is licensed under the terms of the MIT license. ※

Making EIT Data with EIDORS

The goal of this project is to make custom-made EIT images with MATLAB and EIDORS.

What you need to run this codes

Language : Matlab R2021b

Upper version might be okay, but without guarantee. However, I suggest you not to use the lower version.

Add-on : Statistics and Machine Learning Toolbox / Simulink / Partial Differential Equation Toolbox / Image Processing Toolbox / Computer Vision Toolbox

I wrote what I installed when running this project, so please keep in mind that there may be useless add-ons.

Package : EIDORS

Preparation

Before running this code, you have to run specific EIDORS version so that 2D GREIT works. I have no idea which version works 2D GREIT, but Netgen v5.0.0 is necessary.

So, if you want to run it without any hesitation, please DOWNLOAD my personal EIDORS 7z file HERE(Link moves to my personal Google drive). This version works perfectly on this file. If you encountered setting Netgen bin file path on MATLAB cmd after running code, please write Netgen v5.0.0 bin folder path, not v.5.3 folder path.

Definitely, NO virus in my personal 7z file! I strongly bet. And I have no idea why there is v.5.3 folder in 7z file, so ignore it please.

But if you have EIDORS with 2D GREIT runnable version, just download shape_library.mat I uploaded above and paste it.

In this shape_library.mat, I added some pointed models(lct_normal1-6, lct_obese1-3). So if you want to run above .m files, you MUST overwrite "shape_library.mat" I uploaded.

I can't remember what lct stands for is. LOL

Also, I'm not sure that all of .m files saves same result value because of research. So you have to modify end of codes (save part) if you have to.

Results

When you run this codes, they will make the result of EIT images with artificially carved lung elements and the original one. I carved elements of left / right / both lungs with appropriate ratio. You can manage how much elements are carved by manipulating collapseArea inline equation.

I made codes using two methods; FER and GREIT, which has different pros and cons for each method. Or you can use the third method you want by modifying codes yourself.

In my CAENR project, I used these generated EIT images as train / test datasets. You can see the samples on the image at Model architecture paragraph.

I used GREIT methods on my CAENR project, which result was better than FER method and comfortable to use.

Tips for making new thorax shape in EIDORS

1. Find appropriate axial lung CT image.
2. Load image with x-axis as linspace(-1, 1, num of width pixel) and y-axsis as linspace(-height/width, height/width, num of height pixel).

If the size of image is 1000×500, then set x-axis as linspace(-1, 1, 1000) and y-axis as linspace(-0.5, 0.5, 500), which 0.5 = (500/1000).

3. Point the thorax boundary. Recommended number of boundary points is 40 ± 5.
4. Point the both lung boundary. Recommended number of single lung points is 20 ± 3.
5. Save points as [ x1 y1; x2 y2; ... ], then make a structure includes thorax and both lungs boundary points.
6. Go to EIDORS → models → shape_library.mat, and append the structure in shape_library.mat.
7. Load!

[Fig. 1] How to pointing lung on the CT image

This is the simple example of making user defined shape and load it on code.

After finished picking points, let's make a structure name as adult.

Then save thorax boundary points as adult.boundary,
right lung points as adult.right_lung,
and left lung points as adult.left_lung.

Save structure adult in shape_library.mat which exists in the path written above.

If you don't have any idea how to save the points, then see any structures in shape_library.mat.

Use shape_library function to call your adult shape like below;
  
thorax = shape_library('get','adult','boundary');
rlung  = shape_library('get','adult','right_lung');
llung  = shape_library('get','adult','left_lung');

If you called your defined shape but it doesn't work, change the shape in function ng_mk_extruded_model.

When you change smoothing points higher than 50, then it'll work. But be aware; choose just appropriate number. Using large number on smoothing points loads a lot of burden to your computer resources.

shape = { 0,                % height
    {thorax, rlung, llung}, % contours
    [4,50],                 % perform smoothing with 50 points    ← Change here!
    0.04};                  % small maxh (fine mesh)              

elec_pos = [ 16,            % number of elecs per plane
    1,                      % equidistant spacing
    0]';                    % a single z-plane

elec_shape = [0.05,         % radius
    0,                      % circular electrode
    0.01 ]';                % maxh (electrode refinement)

fmdl = ng_mk_extruded_model(shape, elec_pos, elec_shape);

If you cannot figure it out anyway, please see my code.

[Fig. 2] Result of custom generated FEM image from Fig. 1 using EIDORS

Citation

When this codes help your research, I'll be appreciated if you cite this conference paper. This paper is the first announced paper that this custom-made EIT image generation code was used.

Won-Doo Seo and Hyeuknam Kwon, "The Degree of Lung Collapse Estimation Method using Convolutional Autoencoder and Neural Regression in Electrical Impedance Tomography," ICBEM-ICEBI-EIT(International Conference on Bioelectromagnetism, Electrical Bioimpedance, and Electrical Impedance Tomography) 2022, Seoul, Republic of Korea, June 29-July 1, 2022.