Can I create a ground truth for white blood cells by using CellProfiler?
We need some more “ground truth” from you. Can you be more specific?
Yes, I have many images of 5 classes of cells, and I need to segment (manually) the nuclei and cell. Then, I have to get some features from these parts to train a classifier. It posible to do this with CellProfiller?
It is good idea to integrate weka classifiers in CellProfiller?
(1) use CellProfiler’s IdentifyPrimManual to manually outline the objects, however CP is really designed to use a more automatic method like IdentifyPrimAutomatic.
(2) In CP, use lots of measure modules on these objects (MeasureAreaShape, Neighbors, Intensity, whatever you think best, but it doesn’t hurt to measure a lot of features for classification, so be liberal)
(3) Use CellProfilerAnalyst to classify, starting with your ground truth objects, of course. We don’t have a classifier built into Cellprofiler, though these tools are planned to be more integrated eventually. CPA is available from our download site, though it takes more effort to set up than CP.
However, we don’t have methods for using a classifier to alter segmentation, so our standard approach (above) will not alter the segmentation. It will only use the measured features from previously defined, fixed objects. Pixel-based classification is one of our long-term goals to integrate, however you might look to solutions outside CellProfiler if that is really what you need.