A stochastic model for automatic extraction of 3D neuronal morphology
Sreetama Basu1, Maria Kulikova2, Elena Zhizhina3
1National University of Singapore, Singapore.
Medical Image Computing and Computer-Assisted Intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention
|February 8, 2014
Summary
We developed a new method for automatically extracting tubular structures from biomedical images. This approach accurately maps neuronal arbors, including bifurcations and terminals, without user input.
Area of Science:
- Biomedical Imaging
- Computational Neuroscience
- Image Analysis
Background:
- Tubular structures are common in biomedical images.
- Accurate representation of their topology is crucial for analysis.
- Existing methods often require manual interaction or seed points.
Purpose of the Study:
- To introduce a fully automatic framework for extracting tubular structures.
- To enable unsupervised network extraction of neuronal arbors.
- To accurately identify centerlines, width, orientation, bifurcations, and terminals.
Main Methods:
- A stochastic Marked Point Process framework was developed.
- The model uses special configurations of marked objects and an energy function.
- Optimization is achieved via stochastic birth and death dynamics.
Main Results:
- The method successfully extracts centerlines, local width, and orientation of neuronal arbors.
- Critical nodes such as bifurcations and terminals are identified.
- Promising results were obtained on 3D light microscopy images from the DIADEM dataset.
Conclusions:
- The proposed Marked Point Process model offers a fully automatic and unsupervised approach for tubular structure extraction.
- It accurately captures the topology and key features of neuronal arbors.
- The method shows potential for analyzing complex biological networks in 3D imaging data.
