Discrepencies between ROI inspector and ROI statistics measurements

73 views
Skip to first unread message

Igothigh

unread,
Feb 12, 2015, 8:22:54 AM2/12/15
to icy-so...@googlegroups.com
Dear all,

I was double checking fluorescence intensity measurements and surprised to not obtaining consistent results (see screen snapshot). One analysis was done using ROI inspector data (with export to Excel) and the other one by ROI statistics. ROIs are of course the same for each analysis. Checking with ImageJ showed that ROI statistics gave the right measurements (or the same as ImageJ at least...). Unless I missed something, it should be corrected to avoid misinterpretation.
Best.

Stéphane
Clipboard.jpg

Igothigh

unread,
Feb 13, 2015, 9:54:33 AM2/13/15
to icy-so...@googlegroups.com
We checked on Icy vers 1.5.4.2, and both ROI statisitics and inspector gave the same results but different from ImageJ measurements.

Stéphane

Stephane

unread,
Feb 13, 2015, 10:12:38 AM2/13/15
to icy-so...@googlegroups.com
Hi Stéphane,

Do you have the image and the attached xml file so we can test in exact same conditions ? Also the values from ImageJ so we could compare ? We changed something (specially about how we handle channel position set to -1) so it will be nice to test on our WIP version. Thanks !

- Stephane


Le jeudi 12 février 2015 14:22:54 UTC+1, Igothigh a écrit :

Igothigh

unread,
Feb 13, 2015, 11:26:35 AM2/13/15
to icy-so...@googlegroups.com
Here there are. You have an original file (tif + xml) and the corresponding measurements in excel files (inspector, ImageJ and  from a protocol using ROI statisitics).

Stéphane
chromaffin cell bv, NO TRANS +anti OPHN1 48 HRS - Series003.tif
chromaffin cell bv, NO TRANS +anti OPHN1 48 HRS - Series003.tif.xls
chromaffin cell bv, NO TRANS +anti OPHN1 48 HRS - Series003.xml
imageJ results.xls
inspector results.xls

Stephane

unread,
Feb 13, 2015, 12:12:47 PM2/13/15
to icy-so...@googlegroups.com
Wonderful thanks ! I will investigate that in keep you in touch !

Stephane

unread,
Feb 16, 2015, 6:32:47 AM2/16/15
to icy-so...@googlegroups.com
Hi,

Actually we found out from where comes the difference...
It looks like ImageJ is in computing the number of pixel into a mask by using the bounding box of mask instead of counting contained pixel in the mask (as does Icy inspector and should does the ROI Statistics).
So every calculations using the total number of pixels are wrong (as the mean intensity).
For instance you can see that Area or Perimeter are different in Icy and ImageJ.
In ImageJ it's computed with the bounding box : (bb.width * pixelSizeX) * (bb.height * pixelSizeY)
large ROI: (145*68.6) * (144*68.6) = 98.26 µm² ~ 98.4 we have in ImageJ compared to the Icy real area : 74.73 µm²
Actually ROI Statistics introduced a bug lately and should be fixed now !

Best,

- Stephane

Stephane

unread,
Feb 16, 2015, 10:13:42 AM2/16/15
to icy-so...@googlegroups.com
Hopefully the problem is not in ImageJ but just because we use ImageRoi object to convert from the ROIArea in Icy and for some reason the mask is not correctly calculated afterward but we will try to use another ImageJ Roi object to avoid the problem in future ! The important is that both Icy inspector and ROI Statistics will have same and corrects results.

Best,

- Stephane

Igothigh

unread,
Feb 16, 2015, 12:01:02 PM2/16/15
to icy-so...@googlegroups.com
Hi Stéphane,

Thanks for your answer, but despite ROI stat update, I still have different values between ROI stat and the inspector (see screen capture)
Best

Stéphane
Clipboard.tif

Stephane

unread,
Feb 16, 2015, 12:06:57 PM2/16/15
to icy-so...@googlegroups.com
From what Alexandre told me, it will be updated at same time as the next Icy (as it will use some of its news methods).
Don't worry the next Icy should come very soon (i.e. probably tomorrow).

Best,

- Stephane

IcyLyd

unread,
Feb 16, 2015, 12:30:57 PM2/16/15
to icy-so...@googlegroups.com
Great ! , I'm very eager to test it :P
Lydia.

JBD

unread,
Feb 18, 2015, 12:24:59 PM2/18/15
to icy-so...@googlegroups.com
Hi everybody,
I think I might be experiencing the same kind of inconvenience when trying to implement a K means analysis into a basic protocol...
When using K means to threshold my picture I get mean intensities that are, let's say, logical given the ROI thresholds. However when implementing it into a very basic protocol (see below), the analysis gives values for min / avg / max intensities which are nearly similar for each ROI determined by K-means... which is a problem...

(here are the detailed results)



Thanks in advance for your help
Regards
Jean-Baptiste

Alexandre Dufour

unread,
Feb 18, 2015, 12:33:25 PM2/18/15
to icy-so...@googlegroups.com
Dear all,

You might have noticed that Stephane just released moments ago a new Icy update (1.6.1), which should fix these discrepancies (in addition to the usual fixes and improvements).

Could you double check on your various datasets whether this is the case and report back?

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/d696d858-e4f0-40ce-8910-908d8197c58b%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

JBD

unread,
Feb 18, 2015, 12:46:53 PM2/18/15
to icy-so...@googlegroups.com
I checked everything, updated every plugins, and still have the same problem...
I forgot to mention that the "ROI Statistics" Block in my protocol takes like more than 120 sec to run... which may indicate that the problem comes in fact from the protocol itself...
Best
Jean-Baptiste

Igothigh

unread,
Feb 18, 2015, 1:04:14 PM2/18/15
to icy-so...@googlegroups.com
Hi Alexandre,

Nothing has changed. I obtain the same values as before.

Stéphane

Alexandre Dufour

unread,
Feb 18, 2015, 6:43:33 PM2/18/15
to icy-so...@googlegroups.com
Hi again,

I double checked (again) on the image Stephane posted earlier in the thread, and I can confirm that all measurements are now coherent across all tools (ROI panel, ROI statistics, ImageJ measurements), as seen on the attached screenshot.


Could you please double check that you have the latest Icy (1.6.1) as well as the latest ROI Statistics (4.0.6)? And perhaps an(other) sample dataset with ROI and wrong measures?

Cheers,
Alexandre

PS: the long processing time of the ROI Statistics block is probably due to the fact that as of the 4.0.0 version, *many* more parameters are calculated (notably shape measures), which may indeed become noticeable for a large number of ROIs… 


Igothigh

unread,
Feb 19, 2015, 2:55:58 AM2/19/15
to icy-so...@googlegroups.com
Hi Alexandre,

I checked with latest versions and indeed measurements are now coherent between inspector and ROI statistics (and ImageJ). Thanks for your job!!

Best.

Stéphane

IcyLyd

unread,
Feb 19, 2015, 3:19:22 AM2/19/15
to icy-so...@googlegroups.com
Now everything is fine on my side. :)
Thanks a lot.
Lydia.

JBD

unread,
Feb 19, 2015, 5:49:23 AM2/19/15
to icy-so...@googlegroups.com
Same for me, problem solved now!
Thanks a lot!
Jean-Baptiste

Mafalda Pimentel

unread,
Aug 27, 2015, 6:57:55 AM8/27/15
to Icy imaging
Hello everyone. I'm using the spot detector and I noticed the same issue, ROIs have different values when comparing inspector to ROI statistics. I went to check which values were correct directly in the image and the ROI spots match the ROI statistics, not the inspector.

I uploaded a printscreen where you can see for example spot#121.

Mafalda
Capture.PNG

Stephane

unread,
Aug 27, 2015, 7:34:42 AM8/27/15
to Icy imaging
Hello Mafalda,

I guess you're speaking about the X information ?
Actually in the spot detector the X column give you the X mass center position where in the ROI table it gives you the X position of the bounding box (the left starting position). In the next version of Icy, the ROI Panel will distinguish the X position from ROI bounding box and the X center of mass position to avoid confusion !
Hope that helps !

Best,

- Stephane

Mafalda Pimentel

unread,
Aug 27, 2015, 7:44:43 AM8/27/15
to icy-so...@googlegroups.com
Thank you Stephan, I realized that meanwhile while working with nuclei (bigger that spots). Nevertheless, to get area (2D) in the ROI statistics I should look at the interior, right? It also doesn't match with the inspector... (see example in attachment).

Thanks for the help
Mafalda

--
You received this message because you are subscribed to a topic in the Google Groups "Icy imaging" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/icy-software/3TxuClif3tk/unsubscribe.
To unsubscribe from this group and all its topics, send an email to icy-software...@googlegroups.com.

To post to this group, send email to icy-so...@googlegroups.com.
capture2.PNG

Stephane

unread,
Aug 27, 2015, 9:42:10 AM8/27/15
to Icy imaging
Actually in spot detector the interior detector is computed by taking pixel entirely contained in the ROI, it does not count partially contained (intersected) pixel at all. In the ROI panel from Icy we try to give a better approximation by taking in account these partially contained pixels, it's why the value is a bit higher... neither results are really wrong at the end, the important is to compare values on a same calculation method.

- Stephane
Reply all
Reply to author
Forward
0 new messages