Segmentation of cardiac MRI using particle filters

1
Segmentation of cardiac MRI using particle
filters

• Leyla Imanirad, University of Toronto
• Edward S. Rogers Sr. Department of Electrical and
  Computer Engineering
• Institute of Biomaterials Biomedical
  Engineering (IBBME)

2
Outline

• Objective motivation of project
• Problem description
• Methodology
• Particle filters
• Results
• Future directions

3
Objective Motivation

• Automatic segmentation of left ventricle (LV) in
  4D(3D time) MRI
• Segmentation data is used in derivation of
  physiological parameters
• Manual segmentation of 80-120 2D-images is
  time-consuming and error-prone

Figure 1. Cardiac segmentation
4
Problem Description

• Initialize (r , d) on the first line
• Estimate these values for other radial lines in
  the same frame using particle filters

Figure 2. Sample short-axis cardiac image
5
Methodology

• Use gradient values along the line and weighted
  sample set from previous line to construct new
  sample set

6
Particle Filters

• Select particles with larger weights
• Apply motion model to particles
• Update weights based on a likelihood function

Figure 4. Example of different samples
7
Intermediate Results
Figure 5. Intermediate results for frame 8
8
Final Results
9
Future Work

• Improving results by spatial cross-coupling
  between estimated states for each line
• Propagating results to successive frames
• Using more accurate motion model
• Testing with different set of parameters