Académique Documents
Professionnel Documents
Culture Documents
600
the intracranial tissue consist of CSF and cortical gray
matter. Since meninges shares the same gray value as gray
matter, the segmented brain contour is encapsulated with
meninges of varied thickness. Though this global
thresholding produce perfect result for posterior and middle
part of the brain, nasal cavity is misclassified as brain tissue
in frontal area. This non-brain region is strategically
eliminated after identification of the erroneous slices.
Position of the centroid of the bony skull in the brain
bounding box plays a deterministic role in this strategy [28,
29]. Thus, the brain contour surrounded by CSF and
encapsulated by meninges is extracted automatically. One
pixel thick outer boundary of the meninges serves as the con-
tour initialization for cortical reconstruction. End results of
this automated histogram based segmentation spanning the
entire brain from posterior to anterior are depicted in Fig 2.
601
Fig 2: Initial contour: Initial contour for deformable model derived by automated histogram based segmentation
∂ ∂v ∂2 ∂ 2v where S i ∈ {0...1}
w1 − 2 w2 2 − ∇ Ε ext = 0 (3)
∂s ∂s ∂s ∂s The relevant ri ( f ) for the highest Si is considered as the
A numerical solution of (3) is found by discretizing the Rayleigh for the noise distribution rnoise ( f ) . Peak of the
equation and solving the discrete system iteratively. curve rnoise ( f ) coincides with the noise peak of the original
histogram. Then, rnoise ( f ) is scaled by a constant K and a
3.1.1 Anatomical guidance to define external force field
The basic objective of classical deformable model is to bring least square fit is performed by minimizing the error
the initial contour in the area containing CSF. For this expression
2
reason, we have designed an external force field by using fc
f f2
topological guidance of distributed CSF. For this purpose, h( f ) − K exp − (7)
topological distribution of CSF is segmented out. Since we f =0 σ 2
2σ 2
have the image with very distinct pixel values for different to get the best fit Rayleigh curve r ( f ) . Lower threshold of
tissue types, a binary image of CSF and the background has r ( f ) is considered as the lower threshold of the third mode
been produced by automated optimal global thresholding [35,
that is image background.
36] of first two mode of the histogram. Lower threshold of
Part of the histogram below this threshold is clearly
the third mode of the histogram is determined by the
bimodal with well-defined peaks and valley for our image
background removal algorithm [34] as follows:
dataset. First mode is of CSF and the second mode is of gray
In reconstructed MR data, background noise is
matter. Because of this distinct bimodal distribution,
normally distributed white noise and has a Rayleigh
automated optimal global thresholding of the histogram up to
distribution in the image histogram [37] and expressed as
lower threshold of the third mode by the algorithm proposed
2
f f by Otsu [35, 36] efficiently segment out the distributed CSF
Pn o is e ( f ) = exp − (4)
in the given image. Fig. 3 shows an example of segmentation
σ 2
2σ 2
of CSF with intermediate result of lower threshold of the
where σ is the standard deviation of the channel noise. third mode at gray level 146 and optimal global thresholding
Global maxima or height of any Rayleigh curve rmax is at gray level 66. Logic operation of this binary image of CSF
related to σ as σ = 0.607 rmax (5) with the original image produces uniform gray value of zero
All maxima h( fi ) of the histogram function h( f ) and the for entire CSF region instead of a range of gray values.
Thereby it produces a sharp distinct boundary at
corresponding gray values f i are traced. Using this value of meninges/CSF interface and CSF/gray matter interface with
h( fi ) in (5), initially a Rayleigh curve ri ( f ) is generated at enhanced gradient. Now, external potential is designed
whose local minima coincide with step edges and expressed
each maximum. Similarity of ri ( f ) in the range σ i to the cut
Ε
2
off value f c = 2σ i is measured quantitatively with the as ext
( x, y ) = − ∇ I ( x, y ) (8)
original histogram in the same range using a similarity index As meninges has got spatially varied thickness,
derived from the Kappa statistic as proposed by Zijdenbos et depending on the sharpness of the edge gradient and the
al. [32]. If area under ri ( f ) in the mentioned range is chosen parameters for the internal force, it may fail to come
completely inside the meninges. Besides these, in some area,
denoted as Ri and area under the original histogram in the
cortical surface may be attached to the meninges leaving no
same rang is denoted as H i , then the similarity index Si visible subarachnoid space for CSF. Fig. 4 shows the end
becomes result of this step. To overcome these entire problems,
Ri Hi topological guidance along with a prior knowledge of
Si = 2 (6) neuroanatomy has been incorporated in the next step.
Ri + H i
602
by pixel along its length with inspection of surrounding eight
neighbors of each pixel. If any neighbor of the scanned pixel
contains the pixel value zero, it must be part of CSF and the
corresponding CSF set is traced by extraction of connected
component of pixel value zero [38]. If two pixels on the
length of the contour are adjacent to the common CSF set,
connected component of non-zero pixel value in between
them is identified and turned into zero in order to eliminate
those sets. In the thick meningeal fold where the classical
snake could not touch the meninges/CSF interface,
(a) eliminated in this way on letting the cerebrum and
cerebellum the only tissue with non-zero pixel value.
603
g (⋅) gets smaller as h(⋅) becomes larger [4, 30, 31]. For this exactly similar pattern is 1 and index of more than 0.7 is
application, the weighting functions are as follows: considered as very good agreement [41].
2
∇f 5. Conclusion
g ( ∇f ) = exp− (10)
κ In this paper we have tried to meet the basic challenge of
h ( ∇f ) = 1 − g ( ∇f ) (11)
reconstruction of outer cortical surface from MR image of
brain by developing a novel AGHD model which is
The specification of κ determines to some extent the degree hybridization of traditional deformable model with GGVF
of tradeoff between field smoothness and gradient deformable contour along with prior precise anatomical
conformity. knowledge. The proposed algorithm claims its excellence in
Once the GGVF is designed it acts as an external force terms of fully automation, tuning of the least number of
field and replaces the potential force − ∇Eext in (3). Then parameters and avoidance of any assumption or
approximation. No step of the algorithm, from elimination of
the equation is solved numerically by discretization and
bones and soft tissues or initialization of the deformable
iteration, in identical fashion to the traditional deformable
contour to final localization of AGHD model requires an iota
contour.
of human interpretation. Essence of the algorithm resides in
the rich non-overlapping tissue specific information in the
4. Test Results and Validation Checking
acquired datasets. Though the algorithm is specific for the
The proposed algorithm has been tested with 24 datasets of proposed acquisition protocol, the protocol is highly
coronal views of proposed acquisition protocol with excellent applicable for the basic purposes of cortical reconstruction
accuracy. Fig. 5 shows final contour of the AGHD model viz. image registration, image-guided neurosurgery and
overlaid on the original image with intermediate outcome of functional mapping when co-registered with functional
reinitialized deformable model for GGVF (discussed in imaging modalities like SPECT, PET or fMRI since it
section 4.2.2) along with edge map for external force and produces markedly different gray values for different class of
initial estimator. tissue.
For validation study, some slices were selected such In our data sets, though the signal from the gray
that the entire range of the image volume from anterior to matters has been attenuated, it could not be nullified because
posterior was covered. A domain expert traced the target of hardware limitations and unavoidable computational
manually on each slice and it was compared with the constraints [27]. However, the gray-white contrast is
automatically drawn head and brain contour using the markedly improved evidenced by distinct intricate cerebral
similarity index described by Zijdenbos et al.[32]. and cerebellar gray structures. Analysis of the histogram
In case of segmented brain images of the coronal view reestablishes this subjective evaluation. The unique feature of
manually segmented cortical outer contour was highly the histogram is that modal distribution of the gray values
comparable to the automated ones and it is reflected by high directly correlated with a specific anatomical structure. From
similarity indices, approximately 0.98 or more. Index for the histogram one can easily identify a mode actually
representing a given anatomical structure. That is why
Fig. 5: Cortical segmentation: Initial estimator for classical deformable model in first column, edge map for external force
with superimposed reinitialized minimum perimeter polygon (red line) for GGVF deformable model in second column and
final contour of GGVF model with segmentation of CSF/gray matter interface (red line) in last column.
604
determination of the initial contour is easily be made by remains as a set of non-zero value. Polygon of minimum
global thresholding only. Actually, using global thresholding, perimeter of that set serves as initial contour for GGVF
segmentation of the brain is possible only because of high deformable contour and this initialization ensures starting of
contrast of the adjacent tissues as well as full control to the deformable contour in the capture range of the force field.
manipulate it to meet the objective. As in our dataset pixel In our GGVF model, we have chosen a pair of weighting
values of CSF and gray matter are consistently lower than the function for better convergence of boundary concavity which
background noise, lower threshold of the background is is very essential for our application to reconstruct the highly
confidently chosen for the upper threshold for segmentation convoluted cortical surface. The GGVF field computed using
of the initial brain boundary. As the brain is bounded by the this pair of weighting functions will conform to the edge map
bony skull and the background noise never overlaps the gradient at strong edges, but will vary smoothly away from
distribution of the bones, the chance of misclassification of the boundary. Despite medialness of the GGVF deformable
the bony part as brain is virtually eliminated. The only model, in this application the final contour perfectly lies over
shortcoming of the global thresholding is that the nasal cavity CSF/gay matter interface, since the contour faces a very high
is erroneously included in the segmented contour. This has gradient at this interface when approaches externally. This
also been successfully eliminated by incorporating spatial has occurred due to anatomical guidance of CSF distribution.
information of the intricate anatomy [28, 29]. As structural Instead of having a range of gray values, distributed CSF
organization of the bony cranium is more consistent than any posses gray value zero and gray matter distribution in the
soft tissues we intend to shift the centroid towards bony gray scale is far away from zero. This produces very high
region to set a criterion. Sphenoid and the bony orbit help to gradient at their interface and helps to segment out that
identify the anterior slices with coronal section of the frontal interface. The result is highly encouraging and this 2D
lobe of the brain attached to the nasal cavities. No imprecise deformable contour is to be implemented as 3D surface for
morphological operations have been used in this algorithm reconstruction of the entire cortical surface.
since our primary objective is measurement of subcortical
gray matters as well as quantitative study of cognition. 7. References
The initial brain contour derived in this way is
surrounded by CSF and encapsulated by meninges with [1] Isaac N. Bankman, Handbook of Medical Imaging
spatially varying thickness. Meninges is eliminated by using Processing and analysis, Academic Press, 2000
traditional “snake” model and thus the deformable contour is [2] Marijn E. Brummer, Russell M. Mersereau, Robert
brought to the inner boundary of meninges. To design the L. Eisner, & Richard R. J. Lewine, “Automatic
exernal force field we again exploit the unique feature of the Detection of Brain Contours in MRI Datasets”, IEEE
histogram that is modal distribution of the specific tissue Transactions on Medical Imaging, 12(2), pp.153-166,
class for segmentation of topological distribution of CSF in 1993
the image. By assigning the zero value to the distributed CSF [3] Tim McInerney and Demetri Terzopoulos,
map and logic operation with the original image we actually “Deformable models in medical image analysis: a
tries to have zero valued CSF and nonzero valued intracranial survey,” Medical Image Analysis, 1(2), pp. 91-108, 1996
tissues like cerebrum and cerebellum. This step is the most [4] C. Xu, “Deformable Models with Application to
crucial for success of the algorithm. By this way gradient of Human Cerebral Cortex Reconstruction from Magnetic
CSF/meninges interface and CSF/gray matter interface Resonance Images”, Ph.D. thesis, The Johns Hopkins
becomes very high which helps to define the external force University, Baltimore, MD, January 1999
field for traditional snake as well as GGVF field in next step. [5] M. Kass, A. Witkin and D. Terzopoulos, “Snakes:
Besides this, it provides precise anatomical guidance to Active contour models”, International Journal of
eliminate thick meningeal folds which can not be eliminated Computer Vision, 1(4), pp. 321-331, 1988.
by the traditional snakes. Though some zero valued CSF [6] S. Osher and N. Paragios, “Geometric level set
remain distributed in the ventricles, being far away of the methods in imaging, vision and graphics,” Springer-
cortical gray matters they do not affect the following steps of Verlag, 2003
the algorithm anyhow. [7] G. E. Christensen, S. C. Joshi and M. I. Miller,
Final contour of the traditional snake is scanned “Volumetric Transformation of Brain Anatomy”, IEEE
pixel by pixel for its adjacency to the zero-valued CSF. In Transactions on Medical Imaging, 16(6), pp.864-877,
1997
case of any thick isolated meningeal fold, final contour of the [8] Chris A. Davatzikos, “Spatial transformation and
traditional snake passes arbitrarily through mid portion and registration of brain images using elastically deformable
may not be adjacent to CSF. But first and last pixel of the models”, Computer Vision and Image Understanding,
fold must be adjacent to the same CSF set. On checking this 66, pp. 207-222, 1997
condition after tracing these extreme points, non-zero set in [9] W. E. L. Grimson, G. J. Ettinger, T. Kapur, M. E.
Leventon, W. M. W. III and R. Kikinis, “Utilizing
between is identified and turned into zero value to eliminate segmented mri data in image-guided surgery”, Int. J.
it. This situation is more vivid around superior sagittal sinus. Pattern Recog. Artif. Intell., 11(8), pp. 1367-1397, 1996
This checking is made strategically not to affect the part of [10] M. Vaillant, C. Davatzikos, R. H. Taylor and R. N.
the cortical folds which are attached to the meninges leaving Bryan, “A path-planning algorithm for image-guided
no CSF in between them. Thus all discrete non-zero valued neurosurgery”, in Lecture Notes in Comp. Sci.:
CVRMed-MRCAS’97, 1205, pp.467-476, 1997
meningeal folds turns into zero selectively and the brain
605
[11] S. Joshi, J. Wang, M. I. Miller, D. C. V. Essen and Measurement Society, Thessaloniki, Greece, pp. 311-
U. Grenander, “On the differential geometry of the 315, 2010
cortical surface,” in Proc. Of the SPIE: Vision Geometry
IV, 2573, pp. 304 -311, Aug. 1995 [28] Somojit Saha, Sarit k Das, Avijit Kar;
[12] L. D. Griffin, “The intrinsic geometry of the “Segmentation of MR Image of Brain: A New Method
cerebral cortex”, Journal of Theoretical Biology, 166(3), Using Unique Histogram Features and Prior Knowledge
pp. 261-273, 1994 of Brain Anatomy,” IPCV , WORLDCOMP, Las Vegas,
[13] A. M. Dale and M. I. Sereno, “Improved Nevada, USA,2010
localization of cortical activity combining EEG and
MEG with MRI cortical surface reconstruction: A linear [29] Somojit Saha, Sarit k Das, Avijit Kar; “A New
approach”, Journal of Cogn. Neuroscience, 5(2), 162- Segmentation Technique for Brain and Head from High
176, 1993. Resolution MR Image Using Unique Histogram
[14] H. A. Drury and D. C. V. Essen, “Functional Features,” IPTA (IEEE & EURASIP), Paris, France,
specializations in human cerebral cortex analyzed using 2010
the visible man surface-based atlas”, Human Brain
Mapping, 5, pp.233-237, 1997 [30] C. Xu and J. L. Prince. “Snake, shape, and gradient
[15] Chris A. Davatzikos and Jerry L. Prince, “An Active vector flow,” IEEE Transaction on Image Processing,
Contour Model for Mapping the Cortex”, IEEE 7(3), pp. 131-139, 1998
Transactions on Medical Imaging, 14(1), pp.65-80, 1995 [31] C. Xu and J.L. Prince, “Generalized gradient vector
[16] Chris A. Davatzikos and R. N. Bryan, “Using a flow external forces for active contours,” Signal
deformable surface model to obtain a shape Processing, An International Journal, 71(2), pp. 131-
representation of cortex”, IEEE Transactions on Medical 139, 1998
Imaging, 15(6), pp.785-795, 1996 [32] A. Zijdenbos, B. Dawant, R. Margolin, & A.
[17] D. MacDonald, D.Avis and A. C. Evans, “ Multiple Palmer, “Mophometric analysis of white matter lesions
surface identification and matching in magnetic in MR images: Method and Validation”; IEEE
resonance images,” in SPIE Proc. VBC’94, 2359, pp. Transactions on Medical Imaging, 13(4), pp. 716-724,
160-169, 1994 1994
[18] T. Kapur, W. Grimson, W. Wells and R. Kikinis, “ [33] Zhi-Pei Liang, & Paul C. Lauterbur, Principle of
Segmentation of brain tissue from magnetic resonance Magnetic Resonance Imaging – A signal processing
images”, Medical Image Analysis, 1, pp.109-128, 1996 perspective, IEEE Press, 2000
[19] P. C. Teo, G. Sapiro and B. A. Wandell, “Creating [34] Somojit Saha, Sarit k. Das, Subhajit Das, Avijit Kar;
connected representations of cortical gray matter for “An improved algorithm for automated segmentation of
functional MRI visualization”, IEEE Transactions on the head contour from MR image”, VIIP, IASTED
Medical Imaging, 16(6), pp.852-863, 1997 Cambridge, UK, pp. 25-32, 2009
[20] C. Xu, D. L. Pham, M. E. Rettmann, D. N. Yu and [35] N. Otsu “ A threshold selection method from gray
J. L. Prince, “Reconstruction of the human cerebral level histogram”, IEEE transaction on systems, man, and
cortex from magnetic resonance images”, IEEE cybernetics, SMC 9, pp 62-66, 1979
Transactions on Medical Imaging, 18(6), pp.467-480, [36] R. M. Haralick and L. G. Shapiro, “Computer and
1999 Robot Vision, vol. 1, Addison-wesley Publishing
[21] Tim McInerney and Demetri Terzopoulos, “ T – company, 1992
snakes: Topology adaptive snakes”, Medical Image [37] M. Henkelman, “Measurement of signal intensities
Analysis, 4, pp.73-91, 2000 in the presence of noise in MR images”, Medical
[22] X. Zeng, L. H. Staib, R.T. Schultz and J. S. Duncan, Physics, 12(2), pp. 232-233, 1985
“Volumetric Layer segmentation using coupled surface [38] Rafael C. Gonzalez, & Richard E. Woods, Digital
propagation” in Computer vision and Pattern Image Processing 2nd ed. (Prentice-Hall, 2002).
Recognition, pp. 708-715. IEEE Comp. Soc., Los [39] C. E. Kim and J. Sklansky, “Digital and cellular
Alamitos, 1998 convexity,” Pattern recognition, 15(5), pp. 359-367,
[23] X. Zeng, L. H. Staib, R.T. Schultz and J. S. Duncan, 1982
“ Segmentation and measurement of the cortex from 3D [40] J. F. Canny, “A computational approach to edge
MR images using coupled surface propagation”, IEEE detection,” IEEE Transaction on Pattern Analysis and
Transactions on Medical Imaging, 18(10), 1999 Machine Intelligence,” 8(6), pp. 679-698, 1986
[24] R. S. Snell, “ Clinical Neuroanatomy for Medical [41] John J. Bartko, “Measurement and Reliability:
Students,” Lippincott-Raven Publishers, Fourth edition, Statistical Thinking Considerations”; Schizophrenia
1997 Bulletin, 17(3), pp. 483-489, 1991
[25] M J T FitzGerald and Jean Folan-Curran, “ Clinical
Neuroanatomy and Relate Neuroscience,” W.B.
Saunders, Fourth edition, 2002
[26] R.H. S. Carpenter, “Neurophysiology”, Arnold,
Fourth edition, 2003
[27] Somojit Saha, Sarit K. Das, Avijit Kar, “High
Resolution MR Image of Brain With Signal Attenuation
of Gray Matter”; IST, IEEE Instrumentation and
606