3D+t segmentation

[Benjamin Cappe]

Hi,

I'm a beginner in the use of Icy, and it seems that this software can be able to do what I want to do.

However, I'm struggling currently to several point.

I have a 3D stack, with 4 channels (DIC, Hoechst, Draq7, YFP) and time.

The analysis I want to perform is the following:

 * Segmentation of the nucleus

 * Tracking of the nucleus along the time

 * From the nucleus (starting point / primary object) determine a secondary object (cell / cytoplasm) based on YFP signal

 * Tracking of cells along the time

 * Measuring intensity and other parameters cells by cells, along the time

Up-to-now, I adapt a protocol that I took from different one coming from Internet (enclosed).

I'm able to make a correct nuclear segmentation.

However, using 'active contours' in order to determine the cells volume from the nucleus ROI, the level of RAM from the computer increase very rapidly (more than 18Gb use by java once it make the active contours process) and it never reach to end the process.

My personal computer has a 16Gb RAM, but I already tried with a 32Gb RAM, and it gives same trouble: saturation of RAM.

Could someone give me advice on an alternative of the protocol, or advice to better use 'Active Contours'?

Many thanks by advance!

Benji

***

Enclosed are the protocol I design so far, and an example of the experiment I want to analyse (crop to 5 time points) is availlable here: https://www.dropbox.com/s/390nn1u5xzl3bry/benjamincappe_cropt5.tif?dl=0

[Alexandre Dufour]

Hi Benjamin,

Since we don’t have access to your dataset, we can only speculate, but here are 2 quick questions/remarks:

 — How big (in GigaBytes) is your entire 5D time-lapse data? Have you tried running the protocol on a single frame (e.g. by extracting the first frame via “sequence operation > Frane > extract current frame”)?

 — Make sure the metadata (and more specifically the pixel size) of your data set is correct (see the sequence properties tab in the inspector). The X-Y-Z resolution is used by the active contours to determine the sampling precision of each cell, in a parameter called “contour sampling”. In 3D, this parameter reads in microns, and defines the average distance between any 2 consecutive points of the surface mesh generated by the active contours. In your protocol, it is currently set to 2 microns, and I would suggest you set this value to whatever is the Z step of your dataset (it’s a starting point, you can adjust this later).

If you are running into more trouble, I suggest you post here a sample data set (single time point perhaps) so that we can investigate further.

Cheers

Alexandre

--
You received this message because you are subscribed to the Google Groups "Icy imaging" group.
To unsubscribe from this group and stop receiving emails from it, send an email to icy-software...@googlegroups.com.
To post to this group, send email to icy-so...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/icy-software/2d18f0da-7458-48ba-a828-c84c2c39081e%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
<benjamincappe_3D+t_segmentation-1.protocol><benjamincappe_3D+t_segmentation-1.protocol_screenshot.png>

[Benjamin Cappe]

Hi Alexandre,

I know that this thread was open long time ago, but I tried analysis using 2D by other software, but I really need 3D analysis, and Icy seems the best for what I want to do, so I'm back on it.

The 5D timelapse is quite big as 7.11GB (4ch - 13z - 73t - 1002x1004xy - 16bits).

The problem currently is that I'm not able to make cell segmentation based on YFP (channel 0).

I can make segmentation of the nucleus (channel 1) using thresholding and HKmeans.

For the cell segmentation - based on the nucleus, as a starting point - I tried to use Active Contour using nucleus ROIs which were dilated as a start point for Active Contour pluggin.

This is working fine when I have one ROI on 2D (one slice) but seems impossible / crashing / giving no ROI when in 3D.

Enclosed is the protocol I use with a template.

The data I use (2ch [YFP and Hoechst], 13z and 2t) is too heavy to be linked here (100Mo) - you can download it here: Sample 3D+t - 100Mo - 2ch + 13z + 2t

Do you have idea about the best way to perform such 3D segmentation?

Many thanks for helping!

Cheers,

Benji