Question about analysis on deconvolved images

[Avi Jacob]

***** To join, leave or search the confocal microscopy listserv, go to: http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy Post images on http://www.imgur.com and include the link in your posting. *****

Hi all,

Usually I perform analysis on raw data. However, I have a case where without deconvoluton of the confocal acquired images, the thresholding and thus the segmentation becomes very messy. We are looking at 3D volumes and total intensity, of a sub-cellular compartment.

My question is, is it OK to use (Huygens) deconvolved images for subsequent analysis. We are using the Leica built-in 3D analysis, so after deconvolution, we export the images back into the Leica format, and do the analysis there.

Please note, that in this case, we will be comparing cell to cell in the same image, not between images, so I figure that because we are getting ratio's from each image, it should be OK, because I won't have to worry about differences cased by the process itself, between images. I hope I was clear.|

Now, what if I DID want to compare between images?

(A lunch is dependent on the answer...)

Thanks

[Mark Cannell-2]

***** To join, leave or search the confocal microscopy listserv, go to: http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy Post images on http://www.imgur.com and include the link in your posting. ***** I would create a mask based on the deconvoleved data and then integrate intensity of the raw image of the masked region unless i knew that the Huygens processing is energy preserving.

HTH Mark

On 11/09/2016, at 8:55 am, Avi Jacob <[hidden email]> wrote:

Huygens) deconvolved images for subsequent analysis. We are using the Leica built-in 3D analysis, so after deconvolution, we export the images back into the Leica format, and do the analysis there. 

Mark  B. Cannell Ph.D. FRSNZ FISHR
Professor of Cardiac Cell Biology
School of Physiology &  Pharmacology
Faculty of Biomedical Sciences
University of Bristol
Bristol
BS8 1TD UK

[hidden email]

[George McNamara]

***** To join, leave or search the confocal microscopy listserv, go to: http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy Post images on http://www.imgur.com and include the link in your posting. *****

Hi Avi,

As Mark wrote, test if the settings you are using are 'energy conserving'. If not, try to find such settings in lab and if you/your colleagues cannot, ask SVI support for settings.

"Trust, but verify" aka "Trust, verify anyway":

Acquire sub-resolution fluorescent beads and test the claim: A couple of overlapping fields of view. Preferably sparse population of beads and your favorite cells. I suggest 100 nm diameter TetraSpeck

https://www.thermofisher.com/order/catalog/product/T7279

The test procedure and results can become supplemental text or a Nature Methods (open access) or Journal of Biomolecular Techniques, http://jbt.abrf.org/jbt-static/index.cfm/page/jbt_toc.htm  open access and no page charges (check out the first paper of the current issue).

You can also arrange a demo (online training session plus say 1 week trial license) of Microvolution.com, which I know has 'energy conserving' setting (if the computer you want to demo with does not currently have an NVidia GPU, this would be a good excuse - I mean opportunity -- to upgrade to a 1070 or 1080 or Pascal generation TITAN, plus HD 4K monitor). You can also ask MediaCy / AutoQuant for a demo license and 'energy' settings.

Lastly, Leica has a deconvolution option - ask Leica for a temporary demo license (USB dongle) to evaluate from inside Leica software ... may be SVI Huygens or a dumbed down Huygens. I found the deconvolution workflow in the Leica software to be bad. Blast time used was mid 2015 on an SP8, so maybe they've made it into "one click, instant gratification, quantitative", and if not, that should be the goal.

George

On 9/11/2016 2:55 AM, Avi Jacob wrote:

***** To join, leave or search the confocal microscopy listserv, go to: http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy Post images on http://www.imgur.com and include the link in your posting. *****

Hi all,

Usually I perform analysis on raw data. However, I have a case where without deconvoluton of the confocal acquired images, the thresholding and thus the segmentation becomes very messy. We are looking at 3D volumes and total intensity, of a sub-cellular compartment.

My question is, is it OK to use (Huygens) deconvolved images for subsequent analysis. We are using the Leica built-in 3D analysis, so after deconvolution, we export the images back into the Leica format, and do the analysis there.

Please note, that in this case, we will be comparing cell to cell in the same image, not between images, so I figure that because we are getting ratio's from each image, it should be OK, because I won't have to worry about differences cased by the process itself, between images. I hope I was clear.|

Now, what if I DID want to compare between images?

(A lunch is dependent on the answer...)

Thanks

[Vincent Schoonderwoert]

***** To join, leave or search the confocal microscopy listserv, go to: http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy Post images on http://www.imgur.com and include the link in your posting. *****

[Reply from Commercial vendor]


Hello Avi,


As makers of the Huygens software we are happy to reply to your question.


>"Usually I perform analysis on raw data. However, I have a case where without deconvoluton of the confocal acquired images, the thresholding and thus the segmentation becomes very messy. We are looking at 3D volumes and total intensity, of a sub-cellular compartment."


As you mentioned, thresholding and segmentation is often more difficult on (3D) raw data as this data unavoidably suffers from out-of-focus light, noise, and maybe also spherical aberration artefacts. All these imaging issues can be corrected with Huygens deconvolution. Additional imaging artefacts, disturbing proper signal distribution, may be caused by chromatic aberration, crosstalk, and (Z, time) drift (stabilization) issues. These can be corrected with other available Huygens wizard-based tools.
The increase in resolution, and in contrast (signal to noise, and signal to background) that deconvolution offers which can be many-fold, and the additional restoration enables much better definition of the objects and thus better object segmentation.

Another important point to consider is that thresholding and segmentation by the user gives biased results. This is especially a problem if multiple images are compared.
A big advantage of deconvolution is that in many cases the algorithms for automatic thresholding (like e.g. Otsu) provide good results that need no further user interference.

The term 'energy conserving' is not always good with respect to proper image restoration.
Huygens deconvolution will preserve intensity (so intensity quantification/comparison is definitely possible). However, there are some factors that might cause a change in total intensity, for example when there is signal close to the border of the (3D) image, after deconvolution, the signal might be located 'outside' of the actual image volume as this is where the actual object lies. Consequently the signal will be cropped away. You could think of it this way: the actual object is not inside the imaged volume, so this 'loss' in intensity of blurred light is actually beneficial when doing quantitative studies.

>"My question is, is it OK to use (Huygens) deconvolved images for subsequent analysis. We are using the Leica built-in 3D analysis, so after deconvolution, we export the images back into the Leica format, and do the analysis there. "

Signal quantification can best be performed on deconvolved data due to the advantages mentioned above and the local conservation of total intensity.
If you consider two objects being present in your raw image: a large and tiny one. Then the tiny object has more out of focus (blurred) signal outside its real object volume in comparison with the large object where the PSF has more overlap with the real object volume. The tiny object will therfore appear as relatively dim. After deconvolution, the signal within the real object size will increase more for the smaller than the larger object (relatively). The result is a closer match to the original object.

With comparing signal between images you need to take a few things into account to make a good comparison.
See our open wiki website for more information: http://www.svi.nl/RatiometricImages
We have also summarized some important points to consider here http://www.svi.nl/SignalQuantification

As George pointed out: it would be good to calibrate the imaging and processing.

Optimal deconvolution may lead to a significant increase in contrast and thus in the dynamic range of the signal. Consequently, some data scaling might be required. The scaling factor applied is always provided in the Huygens deconvolution wizard (or batch processor). In many cases Huygens will mentions 'no scaling factor is required', which means that you can use the intensities in the resulting image directly for comparison. In case Huygens does mention a scaling factor, then you need to multiply the intensity values in your image with this scaling factor for quantitative comparison.

We always advise our customers to save the deconvolved data as HDF5 format, this will save the actual 32-bit intensity values of the deconvolved result, and will also include any scaling factor if applied. Therefore this format is most straightforward for further quantitative analysis in Huygens.
If you save the result as a 16-bit Tiff, or if you export it to LAS X (also 16-bit), then in most cases a second intensity scaling is required for optimal conversion between 32-bit float and 16-bit. The autoscale to Leica option will stretch the intensity values from 0 to 65535 (16-bit range), so you also need to take this scaling factor into account when comparing different images. The same holds when you save the deconvolved data into any other 16-bit format. When exporting to LAS X, it can be better to use the 'Export to Leica' option (instead of autoscale). The table that is shown in the export dialog window will also mention the scaling factor that is applied to stretch the data. Of course this scaling factor is important to do any intensity quantification between images. For more information about TIFF scaling, please see: http://www.svi.nl/TiffScaling

>"Please note, that in this case, we will be comparing cell to cell in the same image, not between images, so I figure that because we are getting ratio's from each image, it should be OK, because I won't have to worry about differences cased by the process itself, between images. I hope I was clear.|

Now, what if I DID want to compare between images? "

See above.

>"(A lunch is dependent on the answer...)"

For further questions, you are welcome to contact us directly.
Or, even better, discuss details with us during one of our trainings here in Hilversum (The Netherlands). Lunch is included!


Kind regards,

Vincent

On 11-09-16 09:55, Avi Jacob wrote:

***** To join, leave or search the confocal microscopy listserv, go to: http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy Post images on http://www.imgur.com and include the link in your posting. *****

Hi all,

Usually I perform analysis on raw data. However, I have a case where without deconvoluton of the confocal acquired images, the thresholding and thus the segmentation becomes very messy. We are looking at 3D volumes and total intensity, of a sub-cellular compartment.

My question is, is it OK to use (Huygens) deconvolved images for subsequent analysis. We are using the Leica built-in 3D analysis, so after deconvolution, we export the images back into the Leica format, and do the analysis there.

Please note, that in this case, we will be comparing cell to cell in the same image, not between images, so I figure that because we are getting ratio's from each image, it should be OK, because I won't have to worry about differences cased by the process itself, between images. I hope I was clear.|

Now, what if I DID want to compare between images?

(A lunch is dependent on the answer...)

Thanks