Segmentation of 3D Medical Objects by Deformable M-reps

[Last update: by G Tracton]

© Copyright UNC, 2000

Papers

The concepts and methods for this segmentation technique are described in the paper

Pizer SM, Fletcher PT, Fridman Y, Fritsch DS, Gash AG, Glotzer JM, Joshi S, Thall A, Tracton G, Yushkevich P, Chaney EL, (2000). Deformable M-Reps for 3D Medical Image Segmentation. Submitted for publication to MedIA. [Adobe Acrobat Reader PDF format]

Further geometric and proabilistic theory on m-reps can be found in

Pizer SM, Fritsch DS, Yushkevich P, Johnson V, Chaney EL (1999). Segmentation, Registration, and Measurement of Shape Variation via Image Object Shape. IEEE Transactions on Medical Imaging, 18 (10): 851-865.  [Adobe Acrobat Reader PDF format]

Additional Results

These are results beyond those which can be found in the first paper.

Most movies are under construction. Our goal is to finish converting results to movies sometime in Nov-2000.

About movie players for AVI and RealPlayer.

Kidney Results

We are displaying results on kidney images from 6 patients, named kidney 0 through kidney 5. The model was built on kidney 0 and used to segment the other five.The kidney model is a single figure modeling the tissue from the kidney parenchyma to the renal pelvis.

  1. Movie of the rigid registration and atom deformation (segmentation) sequence to extract a kidney from a 3D CT data set:
      RP(), RP(), AVI(), AVI()
  2. Movies of image slices vs. segmented kidneys for results which segmented successfully fully automatically after hand placement of model.

    Stage

    Kidney 2

    Kidney 3

    Kidney 4

    During hand placement the
    user has full control

    RP()
    RP()
    AVI()
    AVI()
    After hand placement

    RP()
    RP()
    AVI()
    AVI()

    		 RP()
    RP()
    AVI()
    AVI()

    After all m-rep stages:
    rigid registration,
    atom deformation

    RP()
    RP()
    AVI()
    AVI()

    After the boundary displacement,
    i.e., at the end of the segmentation

     

  3. Movies of image slices vs. segmented kidneys for results which converged incompletely under automatic operation after hand placement of model  

    Stage

    Kidney 5

    Kidney 1

    After hand placement RP()
    RP()
    AVI()
    AVI()

    Failed initially to segment the kidney adequately:
    rigid registration,
    atom deformation

    		 RP()
    RP()
    AVI()
    AVI()
    [Note how much
    kidney is visible
    outside the model
    at the bottom.]

    After repeating atom deformation
    with the previous automatic result as the model

    		 RP()
    RP()
    AVI()
    AVI()

Cerebral Ventricle Results

We are displaying results on MRI brain images from 2 patients named brain 0 and brain 1. The model was built on brain 0 and used to segment the other one. The cerebral ventricle model is a single figure modeling the tissue from the left lateral first ventricle through the left posterior horn.

Stage

Brain 1

During hand placement the
user has full control.

 RP()
RP()
AVI()
AVI()
[Here, interactive elongation is used along the
figure columns is used in addition to
translation, rotation and scaling.]
After hand placement. RP()
RP()
AVI()
AVI()
[The arrow shows where the posterior horn
was missed by the model.]

After all m-rep stages:
rigid registration,
atom deformation.


[the model fits the posterior horn better.]

Forming a model

TBD