Posted by
Aivar Grislis on
Apr 19, 2011; 6:30pm
URL: http://imagej.273.s1.nabble.com/Whole-slide-image-analysis-with-ImageJ-tp3684908p3684911.html
Hi James,
My project for the Hackathon was for the viewing end of things, to
reduce the memory requirement for large images and provide a smooth
zooming and panning experience.
The existing code is more just a tile server that only displays a fake
image at present. It builds a mip-map pyramid of non-overlapping tiles
(currently 256x256 pixels). Each tile is created on demand from higher
resolution tiles and cached. Currently it only creates 100%, 50%,
25%... zoom levels. The tiles are derived from averaging the pixels, so
four pixels become one. This looks good but creates a slight artifical
sharpness. ImageJ currently scales down by decimation, that is it
discards columns and rows.
For the ImageJDev project we are using Johannes' Fiji ImgLib as our
underlying data model. The usual ArrayContainer is limited to around 2
gigapixel images due to Java's use of integer array indices. There is a
CellContainer that can be used to hold huge images that fit in memory or
could support image tiles that are swapped out.
It's interesting to learn of this work, particularly the WSI file
formats and their streaming WSI server.
Aivar
On 4/19/11 10:14 AM, Deeds, James wrote:
> Hi Johannes,
>
> I agree that it would be great to see this code in the public domain. We will contact Dr. Zerbe and see if he is able to provide his code back to the IJ community.
>
> In our lab we deal exclusively with these 'whole slide' images - we currently have ~60,000 slides scanned. I have been using ImageJ (and NIHImage) for analysis of 'field-of-view' images for the last 15+ years so I am very interested to see development in this area. As I see it, there are perhaps two different tasks when working with this image type: 1) analyzing tiles from the whole slide file with generation of markup tiles (or ideally a reconstructed whole slide markup file) and 2) viewing the multiscale whole slide images and markups. Ideally, both of these functions will find their way into the new versions of ImageJ that the team is hard at work on.
>
> I have tried using the LOCI Bioformats Importer plugin to do task #1 but I think that I am running into memory issues even with the 'Virtual Stacks' option checked. We have been using the vendor's (Aperio) 'Extract Region' tool to generate 5000x5000 pixel jpg tiles to folders and then batch processing the files with ImageJ. One challenge is reviewing the markup tiles after analysis and putting the individual tiles in the context of the whole slide image. Another problem is that the extraction process is extremely slow and we don't have an obvious way to batch extract multiple whole slide files. I would like to hear from the user community if anyone has developed a successful strategy for dealing with these images.
>
> It sounds like Aivar Grislis did some work on dealing with large sized images (at least for viewing) at the Hackathon: "Initial development toward a multi-resolution tiled viewer for IJ2 (Aivar Grislis)" I would be interested in hearing from Aivar about his experience working on this project. From the roadmap, it sound like this particular feature is slated for rel 2.5.
>
> Cheers,
> Jim
>
> -----Original Message-----
> From: Johannes Schindelin [mailto:
[hidden email]]
> Sent: Monday, April 18, 2011 4:57 PM
> To: Deeds, James
> Cc:
[hidden email]
> Subject: Re: Whole slide image analysis with ImageJ
>
> Hi James,
>
> On Mon, 18 Apr 2011, Deeds, James wrote:
>
>> I just ran across the article cited below describing a plugin for
>> viewing and analyzing scanned whole microscope slides using ImageJ. I
>> would be interested in trying the plugin out or hearing feedback from
>> others who may have used the tool. I understand that support for whole
>> slide image formats is planned for the big ImageJ 2.0 project underway.
>> Perhaps Dr. Zerbe has some useful input for the ImageJdev team.
>>
>> Distributed computing in image analysis using open source frameworks and application to image sharpness assessment of histological whole slide images
>> Norman Zerbe, Peter Hufnagl, Karsten Schlüns
>> Diagnostic Pathology 2011, 6(Suppl 1):S16 (30 March 2011)
>>
http://www.diagnosticpathology.org/content/6/S1/S16> Sadly, the requirement of other scientific fields for full disclosure of
> methods (including tools) has not made it quite yet into life sciences.
> That is the only reason I can see why the paper describes a plugin that is
> not available publicly.
>
> Therefore I doubt that you will get much feedback from people who used it.
>
> Unfortunately, this is also the reason why nobody involved in the
> ImageJDev project can form an opinion about, or benefit from, the work
> described in the paper, except by putting an undue burden on the
> developers who are already working very hard on a first beta release of
> ImageJ 2.0.
>
> Maybe you want to contact the corresponding author, and maybe the author
> wants to contribute a bit back to the free software he benefitted from so
> nicely? Maybe there is a chance to avoid reinventing the wheel here? Maybe
> this is a perfect opportunity to stand on each others shoulders?
>
> Ciao,
> Johannes