Checking uncertainty in Ilastik

Looking for ways to analyze your dataset? Need help with a software package?
Computational image analysis of multiplexed Imaging Mass Cytometry acquisitions
Post Reply
monirathhav
Posts: 15
Joined: Sat Feb 24, 2018 6:33 am

Checking uncertainty in Ilastik

Post by monirathhav » Thu Oct 04, 2018 3:57 am

Hi,

Is there a way to quantitatively check uncertainty during pixel classification training in Ilastik besides looking at the cyan color opacity?

Thank you.

Monirath

User avatar
votti
Posts: 21
Joined: Mon Nov 27, 2017 12:46 pm

Re: Checking uncertainty in Ilastik

Post by votti » Mon Oct 08, 2018 4:26 pm

Hi,

One way is to export the uncertainties as images and using your image quantification tool of choice (eg CellProfiler or Ilastik) to quantify the total uncertainty of your images/over all images (eg MeasureImageIntensity using CP and exporting the image table as csv) .
If you track how this changes upon training, you should notice total/average uncertainty to reach a plateau after some training. This would then be an optimally trained classifier (as some pixels will be as ground truth uncertain due to our limited resolution (e.g because of cut thickness) ).

Cheers!

Post Reply