Difference between revisions of "EMSegmenter"

From NAMIC Wiki
Jump to: navigation, search
 
(12 intermediate revisions by the same user not shown)
Line 3: Line 3:
 
Tracing the history of the EMSegmenter...
 
Tracing the history of the EMSegmenter...
  
===1995===
+
===1993-1995===
*Image:
 
*Strengths:
 
*Weaknesses:
 
*Publications:
 
  
===1999===
+
====Development====
*Image:
+
The EM segmenter grew out of a collaboration between Sandy Wells, Ron Kikinis and Martha Shenton in about 1993.  The goal was to get good automatic segmentations of white matter and gray matter from T1 weighted MRI.  The biggest difficulty was the intensity inhomogeneities, or "shading", artifact in the images that was due to the MRI scanner used for research at that time.  The effect of the artifact was that a single threshold could not be used to separate white matter and gray matter. 
*Strengths:
 
*Weaknesses:
 
*Publications:
 
  
 +
Various approaches to the problem were tried, some giving good results, but there were remaining imperfections in the results.
 +
Eventually, we decided to construct an explicit representation of the intensity artifact, and attempt to recover the artifact and the segmentation simultaneously.
  
 +
We chose the Expectation Maximization (EM) algorithm, a statistical estimation method that is used when some data is considered to be
 +
"missing".  The result was an iterative algorithm that alternates between two steps. 
  
== EM with Priors ==
+
In the "E" step, the probability of the tissue label at each voxel is estimated, given the image data and the current estimate of the
* [[IMAGE:Ron-ISBI-07.zip| Here]] are the research related slides - I have sometimes a couple of slides on a topic so you can choose.
 
* The [[Media:EMFeedback3.ppt | Feedback slides]] featuring the transition from Slicer 2 to Slicer 3 were generated by Brad and me.
 
 
 
==sw==
 
 
 
 
 
The EM segmenter grew out of a collaboration with Shenton's group in
 
about 1993.  The goal was to get good automatic segmentations of white
 
matter and gray matter from T1 weighted MRI.  The biggest difficulty
 
was the intensity inhomogeneities, or "shading", artifact in the
 
images.  The effect of the artifact was that a single threshold could
 
not be used to separate white matter and gray matter.  At that time,
 
the MRI scanner used for research at BWH had an annoyingly large
 
shading artifact.
 
 
 
Various approaches to the problem were tried, some giving good
 
results, but there were remaining imperfections in the results.
 
Eventually, we decided to construct an explicit representation of the
 
intensity artifact, and attempt to recover the artifact and the
 
segmentation simultaneously.
 
 
 
We chose the Expectation Maximization (EM) algorithm, a statistical
 
estimation method that is used when some data is considered to be
 
"missing". 
 
 
 
The result was an iterative algoritm that alternates between two
 
steps. 
 
 
 
In the "E" step, the probability of the tissue label at each voxel is
 
estimated, given the image data and the current estimate of the
 
 
intensity artifact.
 
intensity artifact.
  
In the "M" step, the intensity artifact is re-estimated, given the
+
In the "M" step, the intensity artifact is re-estimated, given the image data and current estimate of the tissue label probabilities.
image data and current estimate of the tissue label probabilities.
 
  
The EM segmenter proved to be very robust to shading artifacts,  
+
The EM segmenter proved to be very robust to shading artifacts, but in addition, it was also robust to "inter-scan  
but in addition, it was also robust to "inter-scan  
+
inhomogeneities".  With previous classification approaches to segmentation, "training" was needed on a per-scan basis, because
inhomogeneities".  With previous classification approaches to
 
segmentation, "training" was needed on a per-scan basis, because
 
 
of intensity changes from scan to scan.  
 
of intensity changes from scan to scan.  
  
The EM segmener was the first algorithm that could produce high
+
The EM segmener was the first algorithm that could produce high quality segmentations of white matter and gray matter from MRI,  
quality segmentations of white matter and gray matter from MRI,  
+
with no manual intervention needed on a per case basis. This proved to be very valuable in a large longitudinal study of MS in the  
with no manual intervention needed on a per case basis. This proved
 
to be very valuable in a large longitudinal study of MS in the  
 
 
period 1994 - 1995.
 
period 1994 - 1995.
  
 +
{|
 +
|+ '''Fig 1. EM Segmentation: Cross-Sections of Segmentations (1993)'''
 +
|valign="top"|[[Image:one.jpg|thumb|252px|slice of T1 weighted mr (right temporal lobe has bad "shading")]]
 +
|valign="top"|[[Image:two.jpg|thumb|252px|threshoding result]]
 +
|valign="top"|[[Image:three.jpg|thumb|252px|EM result]]
 +
|}
  
Subsequent developments:
+
{|
 
+
|+ '''Fig 2. EM Segmentation: 3D Rendered Segmentations (1993)'''
Tina Kapur: added MRF models and Mean Field solver
+
|valign="top"|[[Image:four.jpg|thumb|252px|3D view of segmented white matter surface from thresholding]]
 
+
|valign="top"|[[Image:five.jpg|thumb|252px|3D view of segmented white matter surface from EM]]
Kilian Pohl:
+
|}
* Added the use of anatomical atlases of specific brain
 
parts, e.g., the hippocampus : started to be a brain
 
"parcellator"
 
 
 
* Added simultaneous registration of atlas and subject
 
 
 
* developed a hierarchial method for parcellation and
 
validated it on schizophrenia data
 
  
 
* developed a mean-field level-set post-processor that
 
is effective for reducing the effects of noise.
 
  
---
+
{|
 +
|+ '''Fig 3. MS Longitudinal Study (1995)'''
 +
|valign="top"|[[Image:seven.jpg|thumb|252px|longituidinal MS, one subect, segmentaiton result without EM]]
 +
|valign="top"|[[Image:eight.jpg|thumb|252px|Results with EM]]
 +
|}
  
images:
+
{|
[[Image:one.jpg|thumb|252px|slice of T1 weighted mr (right temporal lobe has bad "shading")]]
+
|+ '''Fig 4. Surface Coil EM Segmentations(1995)'''
 +
|valign="top"|[[Image:nine.jpg|thumb|252px|surface coil image]]
 +
|valign="top"|[[Image:ten.jpg|thumb|252px|surface coil image corrected by EM algorithm]]
 +
|}
  
[[Image:two.jpg|thumb|252px|threshoding result]]
+
====Publication====
  
[[Image:three.jpg|thumb|252px|EM result]]
+
Adaptive Segmentation of MRI Data. WM Wells III, WEL Grimson, R Kikinis, FA Jolesz. IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 15, NO. 4, AUGUST 1996
  
[[Image:four.jpg|thumb|252px|3D view of segmented white matter surface from thresholding]]
+
===1995-1998===
[[Image:five.jpg|thumb|252px|3D view of segmented white matter surface from EM]]
 
  
[[Image:seven.jpg|thumb|252px|longituidinal MS, one subect, segmentaiton result without EM]]
+
====Development====
 +
From 1995-1998, MRF models were incorporated into the EM segmenter via a Mean Field Solver.  The resultant segmentations more robust to noise.
  
[[Image:eight.jpg|thumb|252px|" " " with EM]]
+
====Publication====
 +
Enhanced Spatial Priors for Segmentation of Magnetic Resonance Imagery. T. Kapur, W.E.L. Grimson, W. M. Wells III, R. Kikinis, MICCAI, Cambridge, MA, Octobery 1998.
  
[[Image:nine.jpg|thumb|252px|surface coil image]]
+
===2000-2007 ===
 +
From 2000 to now, spatial priors were added to the EM segmenter.  Specifically,
 +
* Added the use of anatomical atlases of specific brain parts, e.g., the hippocampus : started to be a brain "parcellator"
 +
* Added simultaneous registration of atlas and subject
 +
* developed a hierarchial method for parcellation and validated it on schizophrenia data
 +
* developed a mean-field level-set post-processor that is effective for reducing the effects of noise.
  
[[Image:ten.jpg|thumb|252px|surface coil image corrected by EM algorithm]]
 
  
Publications:
+
===Additional Slides===
 +
(From Kilian)
  
#Adaptive Segmentation of MRI Data. WM Wells III, WEL Grimson, R Kikinis, FA Jolesz. IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 15, NO. 4, AUGUST 1996
+
* [[IMAGE:Ron-ISBI-07.zip| Here]] are the research related slides
#Enhanced Spatial Priors for Segmentation of Magnetic Resonance Imagery. T. Kapur, W.E.L. Grimson, W. M. Wells III, R. Kikinis, MICCAI, Cambridge, MA, Octobery 1998.
+
* The [[Media:EMFeedback3.ppt | Feedback slides]] featuring the transition from Slicer 2 to Slicer 3 were generated by Brad and Kilian.

Latest revision as of 16:41, 14 April 2007

Home < EMSegmenter

History

Tracing the history of the EMSegmenter...

1993-1995

Development

The EM segmenter grew out of a collaboration between Sandy Wells, Ron Kikinis and Martha Shenton in about 1993. The goal was to get good automatic segmentations of white matter and gray matter from T1 weighted MRI. The biggest difficulty was the intensity inhomogeneities, or "shading", artifact in the images that was due to the MRI scanner used for research at that time. The effect of the artifact was that a single threshold could not be used to separate white matter and gray matter.

Various approaches to the problem were tried, some giving good results, but there were remaining imperfections in the results. Eventually, we decided to construct an explicit representation of the intensity artifact, and attempt to recover the artifact and the segmentation simultaneously.

We chose the Expectation Maximization (EM) algorithm, a statistical estimation method that is used when some data is considered to be "missing". The result was an iterative algorithm that alternates between two steps.

In the "E" step, the probability of the tissue label at each voxel is estimated, given the image data and the current estimate of the intensity artifact.

In the "M" step, the intensity artifact is re-estimated, given the image data and current estimate of the tissue label probabilities.

The EM segmenter proved to be very robust to shading artifacts, but in addition, it was also robust to "inter-scan inhomogeneities". With previous classification approaches to segmentation, "training" was needed on a per-scan basis, because of intensity changes from scan to scan.

The EM segmener was the first algorithm that could produce high quality segmentations of white matter and gray matter from MRI, with no manual intervention needed on a per case basis. This proved to be very valuable in a large longitudinal study of MS in the period 1994 - 1995.

Fig 1. EM Segmentation: Cross-Sections of Segmentations (1993)
slice of T1 weighted mr (right temporal lobe has bad "shading")
threshoding result
EM result
Fig 2. EM Segmentation: 3D Rendered Segmentations (1993)
3D view of segmented white matter surface from thresholding
3D view of segmented white matter surface from EM


Fig 3. MS Longitudinal Study (1995)
longituidinal MS, one subect, segmentaiton result without EM
Results with EM
Fig 4. Surface Coil EM Segmentations(1995)
surface coil image
surface coil image corrected by EM algorithm

Publication

Adaptive Segmentation of MRI Data. WM Wells III, WEL Grimson, R Kikinis, FA Jolesz. IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 15, NO. 4, AUGUST 1996

1995-1998

Development

From 1995-1998, MRF models were incorporated into the EM segmenter via a Mean Field Solver. The resultant segmentations more robust to noise.

Publication

Enhanced Spatial Priors for Segmentation of Magnetic Resonance Imagery. T. Kapur, W.E.L. Grimson, W. M. Wells III, R. Kikinis, MICCAI, Cambridge, MA, Octobery 1998.

2000-2007

From 2000 to now, spatial priors were added to the EM segmenter. Specifically,

  • Added the use of anatomical atlases of specific brain parts, e.g., the hippocampus : started to be a brain "parcellator"
  • Added simultaneous registration of atlas and subject
  • developed a hierarchial method for parcellation and validated it on schizophrenia data
  • developed a mean-field level-set post-processor that is effective for reducing the effects of noise.


Additional Slides

(From Kilian)