Abstract— Image segmentation
plays a crucial role in image analysis and computer vision which is also regarded as the bottleneck of
the development of image processing technology applications. Medical Resonance Image (MRI) plays an important role in
medical diagnostics and different acquisition modalities are used. Major goal of fMRI data analysis is to recognize activated brain areas
and one of the major steps has segmentation.
ANN is a computational simulation of a biological neural
network, has classified into many networks. Recurrent neural network
specifically in ESNN have implemented fMRI segmentation. The performance of
ESNN for different number of reservoirs, different range of initial weights in
reservoir matrix and different range of initial weights are discussed. For Brain MRI images; features extracted with
ESNN with CC gives 97% accuracy. MATLAB R2011a software was used. The texture features of each class gives high efficiency rate. The quantification of result demonstrates the effectiveness
of the proposed method.
Echo State Neural Network (ESNN), Contextual Clustering (CC), Brain tumor,
Image segmentation plays
an important role in many imaging applications. Medical
Resonance Image (MRI) plays an important role in the field of medical
diagnostics. In medical imaging for
analyzing anatomical structures such as bones, muscles, blood vessels, tissue
types, pathological regions such as cancer, multiple sclerosis lesions and for
dividing an entire image into sub regions such as the white matter (WM), gray
matter (GM) and cerebrospinal fluid (CSF) spaces of the brain automated
delineation of different image components are used. Study and analysis of brain through the images
acquired by various single and multimodalities.
The medical images are captured by different acquisition
modalities like Ultrasounds (US), X-rays, Computed tomography (CT), magnetic
resonance imaging (MRI), Single Photon Emission Tomography (SPECT), Positron
Emission Tomography (PET). MRI can provide plentiful information about human
soft tissues anatomy as well as helps diagnosis of brain tumor.
MR images are used to analyze and study behavior of the
brain. Diseases are shown using MRI signal
between neural activity and the local blood flow that results in BOLD signal.
But are looking at how blood oxygen levels change and assuming that this is
connected to nerves.
Functional magnetic resonance imaging (fMRI) has become an
important method for the investigation of human brain function, both for
research and for clinical purposes. Functional areas identified by motor,
sensory and language tasks have been shown to correspond well with
intra-operative mapping and also with classifically defined anatomical regions
responsible for these functions.
Segmentation of an object in an image is performed either
by locating all pixels or voxels that form its boundary or by identifying them
that belongs to the object. In medical imaging, segmentation is an important
analysis function for which lots of algorithms and methods have been built up.
Segmentation techniques provide flexibility.
The developing platform for the
detection is MATLAB. Introduce to
acquire high-resolution brain images with ultrahigh field (7T) MR scanner and
identify voxels responding to the task using our approach.
need for pre-processing arises from the fact that the raw fMRI data is
contaminated with artifacts primarily due to body movement, Physiological noise
and scanner artifacts during the course of data acquisition. Pre-processing attempts to increase BOLD
contrast and signal to- noise (SNR) in general by removing the amount of noise
as much as possible. Pre-processing tools are SPM, Brain Voyager, AFNI and
etc. The conventional fMRI pre-processing
pipeline includes motion correction, slice-timing correction, co-registration,
Region of interest identification, Bias field correction, filtering and etc.
motion correction ensures the motion-related artifacts in the data is removed
by choosing the reference image volume and realigning all the remaining image
volumes to the reference to minimize the variance caused by motion. A new volume is acquired every repetition
time (TR) in fMRI. During this time period, individual slices in the volume are
acquired either sequentially or in an interleaved manner (where all odd slices
are acquired before even slices). The slices within the same volume are
therefore acquired at different times.
general purpose of filtering is to increase signal-to-noise by reducing
variance in the data. Spatial filtering
or smoothing averages image intensity of one voxel with its neighbors and helps
to reduce the high spatial frequency effects. The
analysis of fMRI data includes two major steps: Pre-processing and Segmentation.
In before pre-processing steps, can acquire the images from the brain using
various steps to be initiated.
variance induced by gross head motion in fMRI time-series represents one of the
most serious confounds of analysis. Before analysis, head motion detection
should be made to evaluate the quality of data. The adjustment may be furthered
by correction based on an estimate from a moving average auto-regression model
of spin-excitation history effects.
implement a voxel-based analysis of imaging data, data from different subjects
must derive from homologous parts of the brain. Spatial transformations are
therefore applied that move and “wrap” the images that they all
conform approximately to some idealized standard brain.
are several advantages of spatial smoothing. First, it generally increases SNR.
The neuropsychological effects of interest are produced by homodynamic changes
that are expressed over spatial scales of several millimeters, whereas noise
usually has higher spatial frequencies. In fMRI the noise can be regarded as
independent for each voxel and has therefore very high spatial frequency
components. Second, it enhances statistical inference.
Image Enhancement and Background Cancelling
remove the random noise and to maintain the boundary information while
producing no additional artifact, the images can be filtered with an
anisotropic filter. There are many black background pixels around MRI of the
brain, which can be removed before the computation since they are not
meaningful for signal calculation or classification. The threshold can be
computed experimentally by one-tenth of the maximum pixel value of a MR image.
segmentation is very useful in separating grey matter, white matter,
cerebrospinal fluid, blood vessels, and other brain structures. Segmentation
methods usually utilize the differences in intensity distribution of different
tissues. An unsupervised approach combines Kohonen self-organizing feature map
and fuzzy c-means can classify brain into six different tissues using
Tl-weighted, T2-weighted and proton density weighted images. In fMRI study,
segmenting these structures helps differentiating functional responses in gray
matter from large vessels. Thus provides better spatial localization and
III. PROPOSED SYSTEM ARCHITECTURE
Segmentation is an important process that helps to
identifying objects in the given image. Existing segmentation methods are not
able to correctly segment the complicated profile of the fMRI accurately.
Segmentation of every pixel in the fMRI correctly helps in proper location of
tumor. The presence of noise and artifacts poses a challenging problem in
This research work proposes a new intelligent
segmentation technique for functional Magnetic Resonance Imaging (fMRI). It is
mainly used for better segmentation of the complicated profile of fMRI. In this
segmentation process, the fMRI image can be segmented with contextual
clustering method and artificial neural networks.
data are analyzed to find the regions with MR signal changes temporally
correlate with the experiment paradigm. Second, a threshold is used to
discriminate the “inactive” brain regions (i.e., those with signal
changes that are more consistent with noise) from the “active”
regions. Finally, the results of the activation analysis are registered to
high-resolution structural images, which are used to more accurately determine
the brain structures involved in the activation task.
inference about specific regional changes requires statistical parametric
mapping. Although statistical methods are capable of identifying the functional
responses from the MR images, they depend on some prior knowledge or assumption
of the physiological response in brain activation. The development of
data-driven post-processing methods capable of identifying unknown response
pattern becomes crucial.
approach provides reliable analysis of the known functional responses. Methods
belonging to this category include: correlation analysis, t-test, general
The functional activities are performed, detection and identification of
tumor in the brain. Preprocessing and
Segmentation is an important
role in order to distinguish between normal patients and their abnormalities or
The proposed method consists of three stages:
A number of online neuroscience databases are available which
provide information regarding gene expression, neurons, macroscopic brain
structure and neurological or psychiatric disorders. The Internet Brain Segmentation Repository (IBSR) provides manually-guided expert segmentation results along
with magnetic resonance brain image data. fMRI slice images have been obtained
from IBSR for use in this research work.
1. FEATURE EXTRACTION MODULE
Feature extraction is essential for segmentation process.
Since the performance of a segmentation algorithm is based on the type of
features used to train the algorithm. In this research work, statistical
features are used to train the proposed segmentation algorithms. This is
achieved by using the contextual clustering (CC) algorithm for feature
extraction and segmentation of fMRI. The 3×3 overlapping windows are used.