Identifying Objects With Only Edge Contrast


#1

I am new to CellProfiler, and am looking for advice to get a basic workflow going.

My goal is to quantify the number of pollen grains, and their size over a series of images.

My problem (I think) is that the IdentifyPrimaryObjects Module is not recognizing pollen grains, because they are being read more like rings than blobs, and often those rings are not 100% complete.

My images were captured with DIC microscopy which provides good edge contrast around each pollen grain, but poor contrast between the center of the grain and the background.

My pipeline, so far, is very simple: LoadImages > Apply Threshold > IdentifyPrimaryObjects.

I have attatched some screenshots, and the sample image I have been working with.

Any guidance greatly appreciated!

note: I had trouble uploading tifs to the forum, so converted to JPG and compressed, therefore the diameter in pixel units set in the IdentifyPrimaryObjects module may seem inaccurate but it was set according to the original tif image.


#2

Hi,
Thresholding generally isn’t a great strategy with brightfield images; using something like EnhanceOrSuppresFeatures “texture” setting (I used a smoothing size of 4 on the image you posted, but as you said it was compressed so adjust accordingly) can help bring out your objects more clearly in an image like this.

Try putting that module upstream of your IdentifyPrimary rather than just ApplyThreshold and give it a try! If that doesn’t work, you could consider using something like ilastik and CellProfiler together to improve the segmentation. Good luck!