3D distance mapping on ImageJ


I am having a problem to measure the diameter of capillaries in 3D on ImageJ, and I need some help please.

I have a stack of images (8-bits) of capillaries that I threshold, on ImageJ, to got a binary stack : image

These images are taken with a light sheet microscope so pixel dimensions are different in x,y and z (0.5, 0.5, 2) µm.

What I want, it is to measure the diameter of these capillaries in 3D on ImageJ. So after some research on internet, I found that there is a pluggin called 3D Suites created by @ThomasBoudier that allows to make 3D Distance Mapping on binary stack and that take in consideration the calibration of voxels.

So, after using the pluggin, I get a EDT stack in 32-bits, that I convert in 8-bits : image

Now, as I want quantitative informations, I analyze the grey-level histogramm of the stack. (Because, on ImageJ, it is said that the euclidean distances after distance mapping are coded into grey values).

The problem is that I don’t understand what I get. The minimum grey value that I get is 7 which should correspond to a 7 µm radius (= 14 µm diameter). However, it is biologically impossible as capillaries have a 3-10µm diameter.

This is what I get : image

So my questions are : using this pluggin, can I get quantitative informations ? if yes, how can I get them ? And using my technique, are the informations obtained with the histogramm really correspond to euclidean distances ? and if yes, are they in voxels or converted in µm ?

Thank you in advance for your help.

Bonjour @NabilNicolas,

Your process is quite right, except for the 32-bits to 8-bits conversion, this conversion is based on your current brightness/contrast adjustment and will modify values. You cannot accurately convert 32-bits values like 1.253 to an 8-bit integer value. If you look at the histogram of the 32-bits EDT stack you will get correct values.

One alternative would be to use localThickness (part of Fiji).




Bonjour @ThomasBoudier,

Thank you very much for your answer.
As I directly do my histogramm analysis on the 32-bits EDT stack (obtained with 3D Suite) , I’m getting something like this :


And these results seem more realistic as most of the counts are associated to radia (bin center) <=5.

Now, I’m just going to test this method on different stacks to see if everything is repetable, and I’ll tick the “Solution” box on your post.

En vous remerciant encore pour votre aide, je vous souhaite une très agréable journée,