In this paper we present a novel approach for brain surfacec characterization based on convexity and concavity analysis of cortical surface mesh. Initially, volumetric Magnetic Resonance Images (MRI) data is processed to generate a discrete representation of cortical surface using low-level segmentation tools and Level-Sets method. Afterward, pipeline procedure for brain characterization/labeling is developed. The first characterization method is based on discrete curvature classification. This is consists on estimating curvature information at each vertex in the cortical surface mesh. The second method is based on transforming the brain surface mesh into Digital Elevation Model (DEM), where each vertex is designed by its space coordinates and geometric measures related to a reference surface. In other word, it consists on analyzing the cortical surface as a topological map or an elevation map where the ridge or crest lines represent cortical gyri and valley lines represents sulci. The experimental results have shown the importance of these characterization methods for the detection of significant details related to the cortical surface.
Segmentation of anatomical structures of the intracranial cavity is a preliminary and main step for the most analysis procedures for brain MRI images. When, the brain surface called also cortical surface is viewed, a human brain appears as a volume with a highly wrinkled boundary surface having numerous long furrows. The term sulci (plural of sulcus) is associated with these furrows and the term gyri (plural of gyrus) designates the regions between the sulci. Geometric modelling, mesh representation and mesh segmentation of anatomical structures in 3D images are becoming an increasingly important processing step. Segmentation of cortical surface should reduce the mesh into meaningful, connected pieces. “Meaningful” implies that the partitioned areas are relevant to the application at hand. In addition, visualization of brain structures such as brain surface, the segmentation allows the automatic identification and labeling of cortical sulci, which will be used in neuronavigation applications, understanding brain anatomy and function, neurosurgeon can easily track the features of interest. Also, segmented sulci from a brain mesh can serve as landmarks, which can be used to register the mesh with other brain meshes to make intra or inter-comparisons. For example, this could serve to measure brain growth and identify diseases.
The most common characterization or labeling of a cortical mesh is into sulcal and gyral regions. The brain gyri can be defined as the top surfaces of the brain folds (Ridges) or as convex regions. The barin sulci can be defined as the area within the brain folds (valleys) or as concave regions. Segmentation of a cortical surface in terms of sulci and gyri can occur in several ways. Classification or partitioning of the cortical surface in concave and convex regions can reduce the size of the Laplacian matrix, in the case of spectral analysis [
In the last several years, many algorithms have been proposed in this growing area, offering several methods. For example, Rettmann’s works [
We propose in this paper a pipeline steps for the problem of 3D mesh characterization of the brain in Magnetic Resonance Images. We have developed two techniques for cortical surface characterizing using discrete curvature computing and Digital Elevation Model (DEM). In the first technique, we segment the obtained surface meshes using a criteria based on discrete curvature. In the second technique, the cortical surface is labeled by transforming the surface mesh into Digital Elevation Model, where each vertex is designed by its space coordinates and geometric measures like orientation or altitude compared to a reference surface. In other word, it consists to analyzing the cortical surface as a topological map or an elevation map where the ridge lines represent cortical gyri and valley lines represents sulci. This last technique is done in three steps: first, extract cortical surface from volumetric brain MRI. Second, compute outer hull surface of brain using a simplified model of Level-Sets method. This outer hull surface will be used as a reference surface. Finally, Digital Elevation Model (DEM) is computed and cortical surface characterization/labeling is done.
This paper is organized as follows: in the next section, we present a 3D brain surface extraction from volumetric MR images using Level-Sets approach. In Section 3, the cortical surface characterization using discrete curvature information is described. In Section 4, we shows cortical surface labeling using Digital Elevation Model. Then, in the Section 5, we present our preliminary results by applying the proposed technique on a MRI images database. Finally, a discussion and a conclusion related to this work are given in Section 6.
Manually segmentation of volumetric images is a complex process which requires lot of time and much concentration to achieve a good quality extraction of regions of interest. For this reason, it is generally interesting to deal with automatic segmentation algorithms. For this purpose, a range of methods including edge based, region based, and knowledge based have been proposed for semiautomatic or automatic detection of various anatomical brain structures. Recently, several attempts have been made to apply deformable models [4-6] to brain image analysis. Indeed, deformable models refer to a large class of computer vision methods and have proved to be successful segmentation techniques for a wide range of applications. Moreover, they constitute an appropriate framework for merging heterogeneous information and they provide a consistent geometrical representation suitable for a surface based analysis.
In some particular Level-Sets [
We propose a method which operates on 3D MRI scans to extract brain surface. First, the data volume is pre-processed with an anisotropic diffusion filtering method [
The aim of the third step is to extract “exactly” the cerebral cortex surface representing the interface between the Gray Matter (GM) and the Cerebrospinal fluid (CSF). For this reason, we propose to use a deformable model algorithm based on the level set technique. We propose also to drive our model by region information instead boundary information, because it is more robust. The process requires an initialization step and speed function. Theoretically, the Level-Set snake is defined as the zero Level-Set of an implicit function defined on the entire volume. This function will change over the time according to the speed term F. The evolution of is defined as in [
The classical speed term is defined as in [16,17]:
Speed term is coupled with the image data through a multiplicative stopping term. The curvature k and the constant force v propagate curve near the region of interest surface.
In this work, we use a simplified version of the Levelsets formulation [
encloses parts of the background, and grows when the boundary is inside the brain region. Here, the speed function usually consists in a combination of two terms: curvature term for smoothness and data term for evolution. The snake evolves using the following equation:
where D is a data term that forces the model to expand or contract toward desirable features in the input 3D MRI data. By making D positive in desired regions or negative in undesired regions. The term k is the means curvature of the surface, which forces the surface to have less area (and remain smooth), and is a free parameter that controls the degree of smoothness.
The speed function depends on the grayscale value input MRI data denoted D at the point x:
where T controls the brightness of the region to be segmented and controls the range of grayscale values around T that could be considered inside the object. A model situated on voxels with grayscale values in the interval will expand to enclose that voxel, whereas a model situated on grayscale values outside that interval will contract to exclude that voxel.
As represented in
Consequently, the three user parameters that need to be specified for the segmentation are T, and. The initial surface obtained after pre-processing must be transformed into a signed distance [
Curvatures can also be used as a height measure. The idea is that ridges and valleys have opposite signed curvatures, and the cortical surface is naturally divided between crest lines or ridges (gyri) and valley (sulci). For this reason, we compute gaussian curvature, mean curvature and the two principal curvatures, and later use them to classify the surface type of vertices. The major complication is that curvature cannot be directly evaluated for triangle meshes because it is mathematically defined for smooth surfaces only. However, discrete differential geometry operators have been developed which can estimate curvatures on traingulated manifolds [19,20]. We apply here some operators, which are derived recently by Meyer et al. [
Mean curvature
Gaussian curvature
As shown in
where
by examining discrete curvatures on triangular meshes, one can achieve the following analysis:
- Concave regions/valleys (sulci): and
- Convex regions/crest lines (gyrus): and
- If the H value is negative, then we have a convex behavior (gyri), otherwise it is concave (sulci).
After the text edit has been completed, the paper is ready for the template. Duplicate the template file by using the Save as command, and use the naming convention prescribed by your journal for the name of your paper. In this newly created file, highlight all of the contents and import your prepared text file. You are now ready to style your paper.
In this second characterization method, we present a pipeline steps for cortical surface mesh labeling. Our
method is based on transforming the brain surface mesh into Digital Elevation Model (DEM), where each vertex is designed by its space coordinated and geometric measures like orientation or altitude related to a reference surface. In other word, it consists to analyzing the cortical surface as a topological map or an elevation map where the ridge lines represent cortical gyri and valley lines represents sulci. The cortical surface characterization process is done in two steps: first, compute outer hull surface of brain using a simplified model of Level-Sets method. This outer hull surface will be used as a reference surface. Finally, Digital Elevation Model (DEM) is computed and cortical surface characterrization/labeling is done.
An outer hull surface, which wraps the brain surface, is computed using a simplified Level-Set model. This model is given as follows:
where V = +1 is a coefficient that controls the speed and direction of deformation (expands). This constant deformation plays the same role as the pressure force. This Levelsets model is implemented using entropy condition originally proposed in the area of interface propagation by Sethian [23,24].
After his stage, we have two discrete meshes surfaces: cortical surface denoted S and outer hull surface denoted as show in Figures 5 and 6.
We have proposed to define three geometric measures for characterizing the cortical surface as shown on
Each vertex of the cortical surface receives the minimum distance that separates it from the outer surface. A vertex from cortical surface will be considered belonging to sulci or gyri according to its height and according to pre-fixed threshold.
The polar angle α is the angle between the normal vector related to vertex P on cortical surface and the vector .
The polar angle γ is the angle between the normal vector related to vertex P on the cortical surface and the normal vector related to vertex Q on the outer hull surface. Where Q is the nearest vertex belonging to the outer surface to vertex P on the cortical surface S.
We have performed a series of experiments on brain MR images from MeDEISA database [
-
-
-
Through this classification process it will be easy to distinguish between gyri that are the surfaces top of the brain folds (ridges), and sulci that are the surface within the brain folds (basins). Each label corresponds to a different color for each geometric measure. An inherent difficultly in the interpretation of brain characterization is that there is no definition of what is correct. Some works use expert points of view from neurologists on where sulci and gyri exist. However, visual results given by Figures 7-9 can be interpreted given the idea that sulci are the surfaces within the brain folds (basins) and gyri are the surfaces at the top of the brain folds (ridges).
This paper proposes novel methods to decompose cortical surface represented by triangle meshes into separate parts based sulci and gyri using geometric measures computed from cortical surface and outer hull surface of the brain. Specially, the first approach use discrete curvature as height in order to distinguish brain sulci from gyru. The second approach is based on transforming the brain surface mesh into Digital Elevation Model (DEM), where each vertex
is designed by its space coordinates and geometric measures like polar angle α, polar angle γ and elevation of vertex d related to a reference surface. After that we will have, a cortical surface as a topological map, when gyri and sulci are very distinguishable and are divided into two different classes.
Visually appealing results can be interpreted given the idea that that sulci are the surfaces within the brain folds (basins) and gyri are the surfaces at the top of the brain folds (ridges). Thus, the prospects of this work would be
to use our approach to address studies on asymmetry of brain anatomy, the inter-individual variability of brain anatomy, neurological dimension of certain mental diseases such as autism and schizophrenia, or, in the context of longitudinal studies on the characterization of brain development for healthy or pathological subject.