Tool for automated pixel annotation followed by manual correction

Hello everybody,

I need to annotate some images to train a neural network to segment my data. Do you know any tool that would allow me to semi-automate this process and then manually correct it? What I need to obtain is a masked stack for every training stack (my data are in 3D) every all pixel of an object, i.e. a nucleus, are labeled with an integer different from other objects (and the background corresponds to 0). To do that, i was using Labkit and manually annotating every nucleus, but it is too much time and effort.
Do you have some advice to speed up this process?

Thank you!!


So… if I were you - I would not lean on manually editing anything. It’s time-consuming and more importantly - it introduces biases into your results. I would use an automatic threshold method (Otsu generally works well for nuclei) - apply that to each of your slices in your stack and create a binary mask (255 nuclei, 0 background). Then you’d be set.

You will have to write a script to work through each slice of an image stack… using the ImageJ1 Macro language is the easiest place to start.

Here are some helpful links to help with the segmentation of nuclei and scripting:

Hi @LucreziaF,

Based on the explanation above, i understand that there is clear different signal (i mean the intensity number) between your nucleus other objects of interest from your background, then no need to do it manually I think. But could confirm this after looking at your image & what you would want to extract from them. You could use CellProfiler too (for 3D images also) which could be completely automated with optimum pipeline. Check out the tutorials here.
Sharing your sample image here might facilitate us to help you better.

Fujifilm Wako Automation (Consultant)
For CellProfiler training or optimised pipeline write to,

Read more on our site.
Yokogawa CV8000 - The Ultimate in Confocal HCS