1

I am trying to implement a 3D segmentation and tracking algorithm of neuron/axon data from fluorescence imaging. I looked into Watershed algorithm, active contours, OTSU thresholding but am unable to make a decision on which algorithm to go for. Further, what is the best way to do tracking across different frames especially since these cell bodies move and sometimes even overlap with each other?

I will be using python libraries to do this. I was thinking of going with Mahotas for the implementation as it has an efficient C++ implementation. Further, I will be using vispy to do the visualization. Thank you very much!

Chits
  • 23
  • 1
  • 4

2 Answers2

0

It is difficult to answer without seeing any reference images. It is not an easy problem so I would recommend looking at research that has been done in the respective areas.

I can recommend http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=5495143 where 3D neurites are segmented using the Hessian in local path finding.

More info on the Hessian for segmenting 3D-structures (e.g., blood vessels in MRI images) can be found here http://liu.diva-portal.org/smash/get/diva2:302939/FULLTEXT01.pdf.

I hope some of this helps.

mags
  • 489
  • 2
  • 11
0

I concur that there is no one "best" algorithm, and what works well for you depends a lot on the kind of data, the kind of artifacts, the required accuracy of the end result, etc.

On algorithm that springs to mind though, since you mentioned tracking, is the Minimum Cost Path method. Its available in scikit-image. It can work well for tracking of elongated structures (wire-like). If you mean tracking in different frames MCP is probably less meaningful.

Almar
  • 4,756
  • 2
  • 13
  • 9