Sphericity segmentation of nuclei in binary segmented images

Ryan_Pollitt · April 29, 2020, 3:42pm

Hi all,

I have recently been working on a project where I analyse segmentations of brain tissue. My question is one that I thought would be very simple. The images are binary and I have been able to measure the sphericity (by first performing a distance transform watershed to make sure touching structures get seperated and to make sure they all get a different color for the sphericity measurement) of the nuclei using MorpholibJ. However, there is also some linear structures in the 3D data, which I would like to exclude by choosing a threshold value for the sphericity and only keeping structures that have a sphericity above this value. This has proven to be much harder than I expected however, so I would like to know if anyone knows a plugin that is capable of doing this.

Thanks in advance

haesleinhuepf · April 29, 2020, 4:22pm

Hi @Ryan_Pollitt,

if you have a label image (where all objects are numbered) and a sphericity measurement, you could use CLIJ2 #clij to kick labels out of your label image. There is a tutorial online, where I do this depending on pixel count. But it should be doable to adapt this for sphericity - again starting point would be sphericity in a table or an array.

before / after:

github.com

clij/clij2-docs/blob/6c1e2e09aa2a596026b1314775919bbf43cd32e6/src/main/macro/superpixel_segmentation.ijm#L81-L98


pixel_count_threshold = 4000;


threshold_vector_and_visualise(pixel_count, labels, pixel_count_threshold);


// This function takes a vector, binarizes it using a threshold and 
// visualises the results as parametric image using a given labelmap
function threshold_vector_and_visualise(vector, labelmap, threshold) {


	// threshold the vector in two vectors:
	Ext.CLIJ2_smallerConstant(vector, small_objects, threshold);
	Ext.CLIJ2_greaterOrEqualConstant(vector, large_objects, threshold);
	// alternative: use binaryNot
	
	// visualise resulting binary images
	Ext.CLIJ2_replaceIntensities(labelmap, small_objects, small_objects_map);
	Ext.CLIJ2_replaceIntensities(labelmap, large_objects, large_objects_map);
	show(small_objects_map, "below threshold");
	show(large_objects_map, "above threshold");

Let me know if you need support getting this work on your data.

Cheers,
Robert

Ryan_Pollitt · April 30, 2020, 7:36am

This really looks like a great solution. However, my Java coding experience is non-existent, so I was really hoping for a magical plugin to save me. Thanks for the help nonetheless!

haesleinhuepf · April 30, 2020, 7:52am

It’s not Java - it’s ImageJ macro. But yes, you may need to code a bit

There will be a webinar about it as part of the NEUBIAS academy next Wednesday. Feel free to join.
http://eubias.org/NEUBIAS/training-schools/neubias-academy-home/

And if you want to get some basic ImageJ macro skill, check out my lecture:

Cheers,
Robert

NML · May 8, 2020, 9:49am

Hi everyone,

On the topic of sphericity, I see that MorphoLibJ calculates it as , and that “normalized such that the value for a ball equals one” (https://imagej.net/MorphoLibJ#Shape_factors).

So, if I get it right, a sphere would have a maximum value of one and anything less spheric than a sphere would have <1. I’m confused since I get a range of values for my objects’ sphericity ranging 0.1 -> ~2.

Shouldn’t 1 be the max possible?