Pipeline to overcome heterogeneity of images


I am trying to create a pipeline with the main focus being quantification of nuclei features in the images of cells found in Pleural Effusion. I have about 100 images that I would like to quantify.I have attached two sample images, one being benign cells and the other being malignant.

The main problem here is that not all images look alike. In some images(benign cells), the nuclei are very prominent, and I am having no problem identifying those nuclei. But in many cases(malignant cells), I am facing difficulties in identifying the nuclei.

What I am looking for, is a single pipeline that can identify the nuclei regardless of the heterogeneity. The pipeline I have attached works fine for the Benign Cells image. But when I try to use it for the malignant cells image, I get terrible results.

Would you have any suggestions as to adding/deleting certain modules that would help overcome the heterogeneity of the images in order to get a single robust that could identify the nuclei in images similar to the ones posted, without the need to change the parameters of each module for each image?

Thank you in advance,
ExamplePipeline.cp (21.1 KB)

Hi Saurabh,

Unfortunately, I don’t think I have much in the way of suggestions for you in this case. IdentifyPrimaryObjects relies on some amount of consistency in terms of detecting objects, so a set of settings that work well for one object size, may not work well for another.