No Positive Pixel Count Detection on some slides for a project

[Foram Vyas]

Hello! 

I'm trying to run positive pixel count throughout a project containing thirteen slides. I have already set the detection parameters and I do not believe them to be extremely constricting. However, when I ran the batch processing script, a few images had no detection present. I tried playing around with the parameters on the images in particular to make sure that it wasn't the parameter thresholds that were the cause. I also tried restarting the program and the computer. I'm not sure what might be the problem. 

As far as I checked, the image resolution and size is comparable throughout the entire project. 

Thank you so much! 

[micros...@gmail.com]

Just to check a few things:

Which version of QuPath are you running (1.2 vs 1.3)?

What are the relative file sizes of the files that didn't work as compared to the ones that did?

What are the sizes of the .qpdata files for each image (or are the ones that failed 0)?

You mentioned playing around with the parameters on individual pictures, does that mean the detection completes fully when they are run one at a time?  Or did those fail as well?

Are the files on a networked folder?

If you have run the batch multiple times, is it always the same images that fail?

[Pete]

My guess is that it's related to this issue: https://github.com/qupath/qupath/issues/67

Which also comes up at https://github.com/qupath/qupath/issues/111

Could this be the reason, or do your experiments with different parameters rule it out?

I hope this issue (and others) will be fixed in the next release with the changes I've described at https://petebankhead.github.io/qupath/2018/03/19/qupath-updates.html#more-useful-output-for-positive-pixel-count-experimental - but I don't use the command much myself and may well be missing other limitations.

[Foram Vyas]

I'm currently using 1.2. 

They're all relative to one another, ranging from about 300 000 - 450 000 KB. 

The sizes don't really differ from which slides work and those that do not. 

The file sizes for the qpdata does differ; the ones that work have from 40 000 - 87 000 KB, but the ones where the detection failed is about 1000 - 2000 KB. 

The detection process does "complete", they do not show any detection though. 

The files are not on a networked folder. 

And yes, it is the same images that fail every time I run the batch.

[micros...@gmail.com]

I would agree with Pete then, it would probably work in 1.3, and the first thing to try would be to lower your Hematoxylin threshold.  In some cases, depending on your image and color vectors, you may even want a negative value there (I would start at -1 and just see if it works), due to the effects of the color vectors.

[Foram Vyas]

Okay! Sorry for getting back to you so late. The Hematoxylin threshold that I was working with was in the negatives; I had it at -2. I will try using 1.3 and hopefully, that helps. Thank you so much!

[Foram Vyas]

Hello! Sorry, it took me a while to get the chance to update to the new QuPath. I tried running the files again and also made sure the image type was set as some images didn't have one set, however, it still didn't work on the same images. I checked the file size and they're also the same as before where the ones that work have from 40 000 - 85 000 KB and the ones that didn't work had about 1000 - 2000 KB. 

How else should I go about to try and fix this? 

Thank you so much for your help!

[micros...@gmail.com]

I would first try tiling and summing the results of the tiles.

You may want to remove parent annotation, depending on how you handle scripting the sum of the data.

If you are using version 0.2.0m2, you may want to wait until the pixel classifier is scriptable and use that, but either way should work.

There are some example scripts here that show how the pixel classifier can be scripted either in 0.1.2 or 0.1.3. They have not been tested for 0.2.0.m2

Original post that resulted in those two scripts found here.

[micros...@gmail.com]

And if you continue to get errors or non-results, please post the error from the Log/View menu (you may want to edit out any personal information or file names).