Multiple hilite / labelling for object classification and artefact detection in segmented cell images


I use KNIME for processing cell images and quantitative data generated in CellProfiler.

To filter the data, I would like to train a Random Forests model to recognise segmentation artefacts, mitotic cells or dead cells within the image sets. In the past, I’ve done this by manually selecting dead cells (objects) using the hilite function within the Interactive Segmentation Viewer, splitting hilited cells off and appending a constant value column to classify them as either dead or living cells. I could then repeat this process for classifying mitotic cells, segmentation artefacts and so on. However, this process is quite lengthy, clunky and manual, ideally I would like to be able to scan the images and classify cells into different groups all at once (rather than hilight, split, classify for each artefact type). It would be great if I could use the hilite function with multiple different hilite groups/colours however I don’t think this is possible.

I wonder if this could be done using the Interactive Labelling Editor instead? I’ve found that when using this node if I select multiple cells and create a new label, e.g. “dead cells” this combines all of those cells in a particular image into a single label rather than classifying each individual label as a “dead cell”. I need Random Forests to run on the single cell objects so this wouldn’t work. I’m not sure of a way to add a new column based on label name regardless (sorry I am still in process of learning KNIME).

Any ideas on what I can do to?

Thanks very much,


Labeling images can contain multiple labels per pixel. The strategy here would be to use indeed one single label for “dead cells”, but also keep a unique label per object. In the Segment Features and Feature Calculator nodes you can then choose to report the overlapping labels, which will give you the information you need for each single object.

Hi @imagejan thanks for replying, I apologise for not checking here earlier. Reporting the overlapping labels definitely feels like a step in the right direction, however, I’ve been noticing some strange behaviours with the Interactive Labelling Editor + Segment Features nodes in producing seemingly random label names for some cell rows and adding new rows. Ideally, I would prefer to not mess with the labels at all as they are important in our workflow for interactivity between images and quantitative single cell data. I’ll keep testing options and playing with the Interactive Labelling Editor node. Thanks very much, I appreciate it.