How load 16-bit images

Hello,
I actually started using Cell Profiler and might need some help.

I want to analyze 16-bit tiff images so I create the following pipeline:
LoadImages
IdentifyPrimaryObjects
IdentifySecondaryObjects

Everytime I load 16-bit images I get an nonsense results even if I uncheck “Rescale intensities”.
If I use 8-bit images is works fine, but for future analysis (measuring intensity, size, …) I want to use the original image type.

Thanks a lot for your help,
Manuela

I used a Nikon Eclipse Ti inverted microscope

I also got the same error using images from a Leika Confocal microscope.
I allready installed the programm new and tried it again, but this leads to the same result.










Hi Manuela,

This is odd indeed. Would you please send an original image so we can inspect it further? (or a set of images from a single site if you have more than one channel). And from what microscope or imaging system did you obtain this?

Thanks,
David

Hi David,
Do you have any idea how to prevend this problem?
Thanks,
Manuela

[quote=“David Logan”]Hi Manuela,

This is odd indeed. Would you please send an original image so we can inspect it further? (or a set of images from a single site if you have more than one channel). And from what microscope or imaging system did you obtain this?

Thanks,
David[/quote]

Hi Manuela,

I looked at the images you posted, and it looks to me like CellProfiler is doing the right thing. The problem appears when you export as 8-bit – the output image and values(!) are scaled to make it look nice, but each image will likely be scaled differently, which is bad. For example, see this screenshot that I took from CP:
cl.ly/3r18192Y2a2y1B3o1Y2o
On the left is one of your 8-bit images, and on the right is the corresponding 16-bit image. CellProfiler scales the image for viewing only (not the underlying data intensities), so that the images look similar. However, the intensity histograms shows differences. The minimum intensity for the 8-bit image is ~0, while the 16-bit image is ~0.004. Many programs will take the minimum intensity in an image and set it to 0 for viewing, but for data analysis this is dangerous. There are also spiky artifacts in the 8-bit image.

In all likelihood, the “nonsense results” you cite are because of a very common situation, namely, the camera is 12-bit and the tiff is saved as 16-bit. Therefore all the pixel values are shoved into a low range from [0,0.0625]. You can use the RescaleIntensity module to correct this if you like, but it just scales all the values up proportionately. As described in the Help text for RescaleIntensity here
cellprofiler.org/CPmanual/Re … nsity.html

Let us know if this helps!
David

Hi David,

thanks for your help!
We are using a 14-bit camera and I tried to resize the images, but I am not so used to analyze microscope images.
Could you do me a hugh favour and upload your pipeline? That might help me a lot for a better understanding how I can load 14-bit images and resize them.

Thanks for your patience.
Manuela

My pipeline is attached. Key points:

  • I unchecked Rescale in LoadImages. For some images, the metadata is not handled properly, including perhaps yours
  • RescaleIntensity – see the Module Notes at the top of the settings page for this module. Basically this remaps the range [0,0.25] to [0,1] for your 14-bit images.
  • The last 4 modules are just there to show that the rescaling does change the intensity measurement, alternately identifying and quantifying the Raw and Rescaled images.

Hope that helps,
David
rescale_example.cp (7.47 KB)

Hi David,
Thanks so much for your help.
Manuela

[quote=“David Logan”]My pipeline is attached. Key points:

  • I unchecked Rescale in LoadImages. For some images, the metadata is not handled properly, including perhaps yours
  • RescaleIntensity – see the Module Notes at the top of the settings page for this module. Basically this remaps the range [0,0.25] to [0,1] for your 14-bit images.
  • The last 4 modules are just there to show that the rescaling does change the intensity measurement, alternately identifying and quantifying the Raw and Rescaled images.

Hope that helps,
David[/quote]