The cells in the “round tester cell trace.pdf” are indeed round in the raw image. However the segmentations are not. Looking at the upper right panel of yours, “Labeled image”, e.g. neither the orange and yellow objects are round. They have large Perimeter-to-Area and thus low Form Factor = (4πArea/Perimeter^2). So I think the issue lies in better segmentation (the “Identify…” modules) and not the Form Factor calculation, per se.
I am attaching a pipeline. Some notes:
- Adaptive thresholding methods are notorious for producing very “rough”, fractal-like patterns, as they propagate out from their seeds. This would increase the Perimeter / Area ratio a lot. I used a “Background Global” thresholding method in IdentifyPrimaryObjects instead.
- The raw images have some “halos” around the perimeter of the objects (are these bright field images?). The segmentation might thus be improved (and thereby removing the spurious “cell” objects) if you first smooth the raw image a little. I added a Smooth module with a Gaussian filter (you can manually adjust the size if needed). This improves the segmentation of the Cells (to my eye) and also minimizes the Airy disk background artifact.
- I added a DisplayDataOnImage so that you can see exactly which objects correspond to which Form Factors. I am getting Form Factors for the 3 cells in your cropped image of 0.35, 0.61, and 0.81. Note that one has a long tail, and another looks like it is actually two cells but one had no Hoechst staining so it did not have an IdentifyPrimaryObjects “seed”. So these lower-than-optimal values make sense. You would need to improve the Hoechst stain to get the latter to seed properly and the other with the long tail, well, just is!
Hope that helps.
round cell tester pipeline_DLogan.cp (7.58 KB)