SEM Image automated image mapping particle analysis

Hi folks,
I am unsure how to proceed with the images to get the best image analysis out of them. The problem is, it’s a series of sem images where again and again single images have an increased background noise (see dropbox link). First I tried to normalize all images using the histogram function/Interactive 3D surface plot to avoid image loss.

Macro Code:
run(“Bio-Formats”, “open=C:/Users/hofmanpa/Desktop/all_Images/Kontrollgruppe/1.1_new.tiff autoscale color_mode=Default rois_import=[ROI manager] view=Hyperstack stack_order=XYCZT”);
run(“Kuwahara Filter”, “sampling=1”);
setAutoThreshold(“RenyiEntropy”);
//run(“Threshold…”);
//setThreshold(0, 117);
setOption(“BlackBackground”, false);
run(“Convert to Mask”);
run(“Fill Holes”);
run(“Erode”);
run(“Dilate”);

Background/Analysis goals

The image shows a polished zirconia surface (streaking in the background) with particles on top (sometimes more, sometimes less). My goal is to obtain all overlying particles in size and shape and to determine the total surface area of the particles.

Challenges

When processing the images, I tried to follow the tips in the video tutorial on FIJI for segmentation.

  1. Preprocessing
    To make the foreground (particles) stand out better from the background (scratches) I tried several filters. Since I want to preserve the particle shapes, I used the Kuwahara filter.
  2. Threshold
    In selecting a suitable thresholding method, through trial and error, I chosed a global thresholding method (RenyiEntropy fits the best)
  3. Clean up
    I tried to improve the segmentation by eroding, dilating and filling holes.

Thanks for your help!

Hi Paul,
Yeah, there is some serious background stuff going on in your images. Is it stitched fields? I guess those lines are from polishing. Happy to chat SEM sample prep if any use to you!

So I resized the first image to 1kx1k to throw away some of the finer detail then did a quick segmentation in Ilastik. It looks pretty good! Is this enough for you?

Oh, I see the later images get a lot worse…
I’ll have a play with those.

Yup, works great on the noisy images too.

Hi John,
thanks for your quick reply! It’s a stitched Image from a dental structure called ‘abutment’. I measured a square field. In addition to the scratches, the surface has varying amounts of yttrium that form artifacts which you measured via ilastik (never tried this program before for segmentation). I’d like to measure the black/dark gray spots ontop. overview_camera_protocol.PNG|234x234

Hi John, I just started to work with FIJI for image analysis. Today i tried Ilastik for the first time - it’s an amazing tool and maybe could solve my problems to identify all objects. While I’m training the program with the different images and my annotations, I thought about how I can save my trained classifier for further segementations of similiar images?
Thanks fpr ypur support!

Yeah, Ilastik is awesome. Just save your Ilastik project to save the classifier. Then you can use the batch processing function to apply it to lots of images :slight_smile: