Shape Analysis for Microscopy

Kangyu Pan

in collaboration with:

Jens Hillebrand, Mani RamaswamiInstitute for Neuroscience

Trinity College Dublin&

Michael J. HigginsIntelligent Polymer Research InstituteUniversity of Wollongong, Australia

Memory Formation

Neuron cells Stimulated synapses

Protein synthesis

Roles of the specific proteins Shape of the synapses

Jens Hillebrand, Mani RamaswamiInstitute for Neuroscience

Trinity College Dublin

Roles of the specific proteins ?

Co-localization of the different proteins

Gaussian Mixture Model

KEY: fitting a GMM to the surface of an object

• directions• distance ?

?

Merge Split

Optimization

Optimized by Split & Merge Expectation Maximization algorithm (SMEM)

Parameters of the Gaussian mixture components

Number of the components

[1] Z. Zhang, C. Chen, J. Sun, and K. L. Chan, “EM algorithms for Gaussian mixtures with split-and-merge operation”, Pattern Recognition, vol. 36, no. 9, pp. 1973–1983, 2003.

Firstly, similar to Zhang’s split technique [1] relied on multiple random splits at each iteration

Publication: K. Pan, A. Kokaram, J. Hillebrand, and M. Ramaswami, “Gaussian mixtures forintensity modelling of spots in microscopy”, IEEE International Symposium on Biomedical Imaging (ISBI), 2010.

Split operationSection(4.2.2)

EM operation

Split Algorithm

Error distribution

Lately, we developed an error-based SMEM (eSMEM) which is deterministic, repeatable, more efficient.

A collection of the error that belongs to each mixture component at each pixel site

Estimation error

Error distribution

|)()(|)( nnormnEMn xIxIxE

)()()( nnmnm xExwxE

From the E-step of EM

New Error-basedSplit algorithm

• directions• distance ?

?

Split2minx

1minx

ij

maxx

minX

j

i

Contour view

)(xInorm

)(xIEM

Results

Publication: K. Pan, J. Hillebrand, M. Ramaswami, and A. Kokaram, “Gaussian mixture models for spots in microscopy using a new split/merge EM algorithm”, IEEE International Conference on Image Processing (ICIP'10) , 3645-3648 (2010).

GUI for the biologists

Co-localization Analysis

Shape of synapses ?

Publication: K. Pan, D. Corrigan, J. Hillebrand, M. Ramaswami, and A. Kokaram, “A Wavelet-Based Bayesian Framework for 3D Object Segmentation in Microscopy”, SPIE BiOS Symposium.

Regeneration of muscle tissue

• Research on a novel technique that uses electrical stimulation to control the growth of muscle cells through conductive polymer materials.

To assess the performance of various processes, we must measure ‘muscle cell density’ quantitatively.Which requires the classification of:

Cell (with only one nucleus)&

Fibres (with multiple nuclei inside cell body)

Michael J. HigginsIntelligent Polymer Research InstituteUniversity of Wollongong, Australia

Skeletal muscle cells & fibres

Cell body(segmentation of the overlapped cell bodies)

Nuclei(Using GMM and optimized with eSMEM)

Skeletal cells & fibres

The number of nuclei in each cell/fibre

Segmentation of the cell/fibre (especially the overlapped cells and fibres)

A NEW ACTIVE CONTOUR TECHNIQUE FOR CELL/FIBRE SEGMENTATIONCellsnake :

Publication: K. Pan, A. Kokaram , K. Gilmore , M. J. Higgins , R. Kapsa and G. G. Wallace, “Cellsnake: A new active contour technique for cell/fibre segmentation”, IEEE International Conference on Image Processing (ICIP'11) , 3645-3648 (2011).

Future work

Organize the algorithms as plug-in tools for the software that the biologists used (like ‘IGOR Pro’).

Run more experiments to further examine the performance of the techniques and submit the dissertation in April.