Segmentation and quantitative evaluation of brain MRI data with a multi-phase three-dimensional implicit deformable model

1
Segmentation and quantitative evaluation of brain
MRI data with a multi-phase three-dimensional
implicit deformable model

• Elsa D. Angelini, Ting Song, Brett D. Mensh,
  Andrew Laine.
• The Heffner Biomedical Imaging Lab
• Department of Biomedical Engineering, Columbia
  University, New York, NY, U.S.A.

2
Overview

• Introduction
• Multi-phase segmentation method
• Random initialization scheme
• Post-processing method
• Experiment data and results
• Conclusion

3
Method
Active contours without edges Chan-Vese IEEE TMI
2001
Segmentation is performed via minimization of
an energy functional derived from the
Mumford-Shah model.

• u0 is the original image, formed by two
  regions of approximately piecewise constant
  intensities.
• C is a curve defined on the image.
• (c1 , c2) are the average intensity values of
  u0 inside and outside C.

4
Method
Extension to Multi-phases
Two phi functions gt Four phases
One phi function gt Two phases
5
Method
Advantages of Method vs. Parametric Deformable
Models

• Arbitrary initialization.
• Topology unconstrained.
• Simple 3D implementation

6
Initialization
Corresponding partitioning of the image domain
into four phases defined by the overlap of the
two level set functions.
Original MRI slice with initialization circles.

• Random initialization ensures robustness of the
  method to
• Variation of user expertise
• Biased a priori information
• Errors in input information influenced by
  variations in image quality.

7
Post-processing
CSF before dilation
CSF after dilation
8
Data

• Clinical data. 256x256x73, 3mm slice thickness
  and 0.86mm in-plane resolution.
• Labeled data correspond to 40h of expert rater
  for 3 normal patients.

• Phantom data. It is convenient to test and
  validate the algorithm.

9
Data
WM
GM
CSF
Average values
Gaussian fit curves
10
Results Phantom

• Error measurements for segmentation of MRI
  phantom

11
Results Clinical

• Error measurements for segmentation of
  clinical MRI datasets

12
Results

• Three dimensional rendering of segmented
  volumes for three cases

13
Introduction

• A four-phase three-dimensional active contour
  method is implemented with a level set framework
  for automated segmentation of brain MRIs.
• An optimal partitioning of three-dimensional data
  based on homogeneity is used to extract different
  tissue types in the brain.
• Experiments on three MRI brain data sets showed
  that the method can accurately identify white
  matter, gray matter and cerebrospinal fluid in
  the ventricles.
• Quantitative evaluation of the segmentation was
  performed.

14
Conclusion

• A novel clinical application and quantitative
  evaluation of a multiphase level-set segmentation
  algorithm is illustrated.
• The segmentation algorithm performs an optimal
  partitioning of a 3-D data set based on
  homogeneity measures.
• Experimentation on three MRI brain data sets
  showed that the method can accurately identify
  regions of WM, GM and CSF.
• Random seeds for initialization was used to speed
  up the numerical calculation .
• Future work will incorporate available
  co-registered PET data to improve the
  segmentation performance by running the algorithm
  on vectorial-type data.