This is a pretty big task, and I have several work-arounds, but I am looking for an efficient way to go from PAS stained slides of lung tissue to defined ROI's over the whole slide (presumably via BLOB detection). I am hoping to get every airway epithelium annotated as an ROI so I can get colorimetric/masking data from the ROI.

That's the simply stated goal, but there are several issues along the way. The imager we are using produces files in the MRXS format. Satyanarayanan group's work at OpenSlide has been great in the initial hurtles of exporting the files into very large tiled TIFF files (~2-5GB each, under JPEG compression, generating a few hundred a week). This has it's own bugs, but the mailing list has been solid.

Given a very large series of very large tiled TIFF files, what is the best way to train a BLOB (or other) detection method to find an ROI? Using small images, I've gotten acceptable mapping with WIKA, but every time I try on larger images I run into errors and corruptions (when I don't run into memory limits). Is ImageJ the best choice for this? If so which plugins would you recommend trying? I've used what comes FIJI and I've been exploring over others. LOCI will load ~80% of the image correctly then run into tiling problems. Given that you can't choose the ROI strictly on color, but human's can quickly ID it, I think a computer will be able to.

I have two systems I'm trying this on:

64-bit Ubuntu, 32GB RAM, FX-9590 processor 
64-bit Windows 7, 16GB RAM, 2x E5430 processor

The Win7 machine is the one required for the imager, and I'm trying to do the analysis on the more powerful Ubuntu box. Deroulers et al produced an interesting work on the subject, however, my group hasn't been able to implement the system as of yet.

I would prefer an open solution, but a proven solution with propitiatory software would also be acceptable.

I thought the over-arching question would fit here, but if people think I should break it up and put smaller questions on SO, I will. I'm just not sure we're tackling this the best way from slide to graph.

  • $\begingroup$ Two general comments: using JPEG compression is lossy, why not use lossless LZW or similar with TIFF? Have a look at Fiji (fiji.sc) the updated brother to ImageJ. It has macro and programming language inputs, a huge user base, is open source, and is designed for image manipulation tasks. $\endgroup$
    – user560
    Commented Nov 9, 2013 at 16:06
  • $\begingroup$ @leonardo As mentioned in the question, I have tried FIJI. Unfortunately I don't have a method that works to go from MRXS to an LZW compressed TIFF. I can go from MRXS to an uncompressed TIFF (5-10GB), then from that to an LZW compressed file, but that is a huge time hole. If you look at the paper, it's a known problem in the ImageJ world, and I think one that has no coordinated effort to solve. I was wondering if anyone else here has tried to tackle this, and we may end up publishing a methods paper if we figure it out. $\endgroup$
    – Atl LED
    Commented Nov 14, 2013 at 17:47


You must log in to answer this question.

Browse other questions tagged .