It sounds like you already have the answer - to import your data using a .vox model and materials file. That's how it works in CST. But take a step back and look at your data. Are you really saying that every single voxel behaves as a significantly different material? I think that is unlikely. You can make a pretty good approximation to the behaviour of a real body by modelling only air/muscle/fat/bone. Why would you expect e.g. fat in one part of the breast to look so different to fat in another part?

You don't say how you derived your voxel tissue properties. Maybe they come from an actual measurement, rather than the more common segmentation and assignment from a lookup table? In which case you might think about using a clustering algorithm on your Debye parameters to reduce the number of different tissue types modelled.

Arthur W Magill : I really thank you very much for your response.

Regarding the breast phantom which I use: It is available on the online phantom repository of the university of Wisconsin Madison on this link:

https://uwcem.ece.wisc.edu/phantomRepository.html

On this link, a number of realistic phantoms for breast with different density classifications exist.

Why I say that each voxel has its properties? This is what I explain below:

In the instruction manual of these phantoms {available on the same link above}, they say that each phantom is represented by two files of data: a "material type file" and the so-called "p value file ". The "material type" file determines the type of the material inside each specific voxel {the type of the tissue} while the" p value" file determines the dispersion curve of that voxel. That is, they consider each tissue type {like "fat" for example} not represented by a single specific dispersion curve. They represent each tissue type as being a "whole area between two limiting curves of dispersion". This means that for fatty tissue for example, we have two curves to determine its dispersion, upper curve and lower curve. Then to assign the dispersive properties to a fatty voxel they use the equation:

dispersion = p*upper curve + (1-p)*lower curve.

Now each voxel has its m value and p value. We look at m value to determine the upper and lower curves (the tissue type is the whole region in between) and use p value in the equation above to get the dispersion curve for that voxel.

I hope you notice that: I didn't say in the question that each voxel is a separate material. I said that each voxel has its own dispersive curve. The voxels for fatty tissue for example are of same type, but with different disperive curves, with the possibility of some voxels to be identical of course.

Regarding the derivation of these phantoms, they say that they are MRI derived phantom. I think they used large number of MRI images of actual clinical MRI imaging of some volunteers. More information on derivation of these phantoms can be found in this online available article:

https://pulse.embs.org/july-2017/human-breast-phantoms/

Or this article

Article MRI-Derived 3-D-Printed Breast Phantom for Microwave Breast ...

I tried using CST VBA methods, but this didn't help. We can write VBA macro to make each tissue type has its dispersive curve, but I couldn't make each voxel has its dispersive curve.

Can you please refer to some "good and compact" reference for that "clustering algorithm"? I didn't touch that technique before.

Thanks very much for the link - that is extremely useful for my own research! I'll read the article to better understand their model and get back to you. But for a first question, what do you want to achieve? Are you trying to make a 'more realistic' EM simulation by taking tissue variation into account? For each tissue type, I could imagine taking a random (or maybe carefully selected) set of samples across the dispersion variation N_disp, such that N_disp

Arthur W Magill : Regarding what I want to achieve: Yes, I want to make the EM simulation as realistic as possible. The application I am working on is a complete breast imaging system using microwaves. The imaging doesn't aim only to detect cancers but also get a good and reliable 3D image of the breast interior, for density classification, estimating cancer risk and monitoring treatments...etc. To optimize such system before turning to a practical prototype, the simulation should be as realistic as possible.

I understand your suggestion and it seems good and applicable in CST, but I want to clarify some points:

1- From the "p" value data of any phantom on their site, I could extract 10^5 different values of that parameter "p". This means we have "at least" 10^5 different dispersive properties. I say "at least" because I didn't account for "m data". We can have two voxels of same value of "p" and different values of "m" which increases the number of different voxels more above 10^5.

2- According to my experience, CST can support "materials file" up to 256 different materials included. If we want to apply your suggestion, I think we will be highly reducing the model accuracy.

3- I need to keep the model accuracy as at is, for the purpose of comparison of my research results with others.

So, I still need to keep the model as it is, even if I could import a reduced accuracy model in CST and complete the process. The aim is not to just get an accepted approximation to the breast into the simulation, but a totally realistic one.

Hi Mohammad Ali, sorry for taking so long to reply.

Yes, you are correct that CST can 'only' represent 256 different materials (I guess that sounded like a lot when CST started, and generally it still is). From a (too) quick read of the UWCEM manual, it looks like the breast model is essentially divided into 7 material types with p and m values, plus 3 others (muscle, skin immersion medium).

So, from the 256 possible values, subtract 4 for muscle/skin/free space/PEC, leaving 252. You can then divide these remaining available materials in different ways. For example, you could divide them equally across all tissues (252/7 = 36 samples/tissue). That means you can take 36 different sigma/epsilon combinations across the range described by p and m, either uniformly or according to some other scheme, giving 36 different defined materials in CST per voxel classification. In the 3D voxel model, sssign these 36 different material types randomly across the region (e.g. you define 36 slightly different versions of fatty-1, same for fatty-2, etc). You might consider building several versions of the model here, with different random selections of sigma and epsilon, and different spatial arrangements of the defined materials.

Compare the simulation results when you use 7 defined materials, and 256 defined materials. This is an important step - simulation is about abstracting away all the unnecessary detail to capture the essential features of your problem. Possibly the sigma and epsilon variations really don't make any difference. Then you know. Possibly they're very important. I'm really interested to know if it makes any (significant) difference. It's (similar to) a question I've often asked myself when modelling RF coils for MRI. Simulation results always show stronger spatial interference in the RF magnetic fields in body models than measured results. I've often wondered if this is due to the artificial local homogeneity of the model - maybe you'll have the answer for me ;-)

I hope this is clear (ask if it isn't), and do let me know how it works out.