Posted by
Yongqiang Chen on
Nov 08, 2019; 7:44am
URL: http://imagej.273.s1.nabble.com/Calculate-the-concentration-in-a-CT-image-tp5022621p5022624.html
Hi Kenneth and Stein,
Thank you for your kind replies.
To summarize the problem in one sentence, it is to replace the CT number with concentration value or whatever else we want.
I believe interpolation can work. Could you please give me some hints for interpolation implement in ImageJ (or to replace CT number with concentration.)?
Thank you so much.
Yongqiang
-----Original Message-----
From: Stein Rørvik <
[hidden email]>
Sent: 08 November 2019 01:54
To:
[hidden email]
Subject: Re: Calculate the concentration in a CT image
Yes this is doable as per Kenneth's explanation, but you really need to understand the physics involved in your imaging.
If you are using monochromatic X-rays (from a synchrotron) and your software's "CT number" is proportional to your material's linear attenuation coefficient, then the simple linear interpolation "method a" will work. As this is quite unlikely (you probably have a polychromatic source if you are using an in-house CT system), it greatly depends on whether or not your elements distribution is radially symmetric in all directions. The reason for this is that the X-rays loose energy as they pass through the sample, so the apparent concentration in the central parts will be lower than in the outer parts. This phenomenon is called "beam hardening". You can compensate this with X-Ray filtering, or software corrections (if the distribution is symmetric). My experience is that software corrections work better, as filtering really kills the signal noise ratio. Anyway your CT software must still be calculating the attenuation coefficients correctly, and that greatly depends on how clever your software is.
This is in any case easy to check: Just scan a round plastic container with pure KI brine, and then check the radial profile. Is it flat? If yes, then "method b" should work; it probably needs an exponential profile. If it is not flat, you must revise your imaging setup and/or your CT software processing settings. If the KI profile is flat, repeat the experiment with pure water. If that is flat too, even "method a" should work. If KI is your heaviest compound in your system and the overall concentration is low, and all your other elements are light (low atomic numbers), then "method a" might also work even if the test profiles for KI is not flat. But again that will depend on your elements' distribution in the sample and the symmetry of everything. So again you need to understand the X-Ray imaging physics and how it applies to your sample.
It will help if you post some example images, including all instrument related metadata. Also please tell what instrument and what software you are using for the generation of the CT images. I have considerable experience with such measurements from µCT images so I should be able to see if your images are analyzable or not.
Stein
-----Original Message-----
From: ImageJ Interest Group <
[hidden email]> On Behalf Of Kenneth Sloan
Sent: 7. november 2019 21:35
To:
[hidden email]
Subject: Re: Calculate the concentration in a CT image
Knowing nothing about this particular application, I will assume nothing.
So far, you seem to have two pairs of (CT Number, concentration):
(loCT, loConcentration)
(hiCT, hiConcentration)
You want to convert intermediate (or more extreme?) CT numbers into concentrations.
Method a) assume everything is linear, and just use linear interpolation:
u = (CT-loCT) / (hiCT-loCT)
CT = loConcentration + (u * (hiConcentration-loConcentration))
I DO NOT RECOMMEND this method!
Method b) gather more data - measure CT numbers for known concentrations, fit a function to your measurements,
and evaluate this function at new measured CT numbers.
I might start by measuring the CT number for (loConcentration + hiConcentration)/2. If (by some miracle) this turns out to be = (loCT + hiCT) / 2, then go to Method a) Otherwise, measure at ¼ and ¾. Continue sub-dividing until you are happy with a low-degree polynomial that fits your data.
The usual caveats about fitting a predictor to data apply - in particular, it is dangerous to EXTRAPOLATE from the measured values. Interpolation is much safer. And, avoid high-degree polynomials which may overfit your data.
Also consider exponential/log functions. If you know anything about the physics of the imaging, use that.
--
Kenneth Sloan
[hidden email]
Vision is the art of seeing what is invisible to others.
>
>
> I am processing a CT image. They have two different brines mixed together. One fluid is high CT number. Its concentration is 2mol/L (KI doped). Another brine is pure water and its CT number is low.
>
> I want to relate the CT number to the concentration value. (Let's
> assume that the brightest part is 2 mol/L fluid and pure water is
> lowest CT number)
>
> Can anybody help me with the above question?
>
--
ImageJ mailing list:
https://eur03.safelinks.protection.outlook.com/?url=http%3A%2F%2Fimagej.nih.gov%2Fij%2Flist.html&data=02%7C01%7Cstein.rorvik%40sintef.no%7Cf22112f0eb0844a6569708d763c37804%7Ce1f00f39604145b0b309e0210d8b32af%7C1%7C0%7C637087563501115145&sdata=XFl%2FV1LOQkf2E3QxibNVnO32vSvvk38lss%2Bs07lT3w8%3D&reserved=0--
ImageJ mailing list:
http://imagej.nih.gov/ij/list.html--
ImageJ mailing list:
http://imagej.nih.gov/ij/list.html