Blood Vessel Segmentation in Complex-Valued Magnetic Resonance Images With Snake Active Contour Model

International Journal of E-Health and Medical Communications, 1(1), 41-52, January-March 2010 41
Blood Vessel segmentation in Complex-Valued Magnetic resonance images with snake active Contour Model
Astri Handayani, Institut Teknologi Bandung, Indonesia Andriyan B. Suksmono, Institut Teknologi Bandung, Indonesia Tati L. R. Mengko, Institut Teknologi Bandung, Indonesia Akira Hirose, University of Tokyo, Japan
aBstraCt
Accurate blood vessel segmentation plays a crucial role in non-invasive blood flow velocity measurement based on complex-valued magnetic resonance images. We propose a specific snake active contour modelbased blood vessel segmentation framework for complex-valued magnetic resonance images. The proposed framework combines both magnitude and phase information from a complex-valued image representation to obtain an optimum segmentation result. Magnitude information of the complex-valued image provides a structural localization of the target object, while phase information identifies the existence of flowing matters within the object. Snake active contour model, which models the segmentation procedure as a force-balancing physical system, is being adopted as a framework for this work due to its interactive, dynamic, and customizable characteristics. Two snake-based segmentation models are developed to produce a more accurate segmentation result, namely the Model-constrained Gradient Vector Flow-snake (MC GVF-snake) and Stochastic-snake. MC GVF-snake elaborates a prior knowledge on common physical structure of the target object to restrict and guide the segmentation mechanism, while Stochastic-snake implements the simulated annealing stochastic procedure to produce improved segmentation accuracy. The developed segmentation framework has been evaluated on actual complex-valued MRI images, both in noise-free and noisy simulated conditions. Evaluation results indicate that both of the developed algorithms give an improved segmentation performance as well as increased robustness, in comparison to the conventional snake algorithm. Keywords: Blood Vessel, Phase-Contrast Magnetic Resonance Imaging, Segmentation, Simulated Annealing, Snake Active Contour Model
introduCtion
Magnetic Resonance Imaging (MRI) has played increasing significant roles in today medical
DOI: 10.4018/jehmc.2010010104
practice. Among the most prominent clinical strength of MRI are its non-hazardous nature and its ability to provide high contrast between various types of tissue. Nowadays the application of MRI has expanded beyond conventional structural mapping into chemically-specified
Copyright 2010, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
42 International Journal of E-Health and Medical Communications, 1(1), 41-52, January-March 2010
and flow-related imaging. Advanced applications of MRI require supports of sophisticated signal processing procedure to infer the desired information from typical Magnetic Resonance images. This chapter presents the development of blood vessel segmentation algorithms dedicated to support the quantitative flow imaging based on Phase Contrast Magnetic Resonance Angiography (PC-MRA). PC-MRA requires the consideration of both the magnitude and phase information derived from the MRI natural complex-valued signals, rather than relies solely on the magnitude information as commonly practiced in the MRI application for structural mapping. MRI is physically based on the Nuclear Magnetic Resonance (NMR) phenomenon discovered at 1946 by Felix Bloch and Edward Purcell. This phenomenon notes the resonance of magnetic systems when triggered with radio waves with frequencies corresponding to its natural magnetic frequencies; that is the gyroscopic precession frequency of the magnetic moment of the nuclei under the influence of an external static magnetic field. By observing the externally measured NMR signals, MRI produces images of the internal physical and chemical characteristics of an object. This is made possible by the spatial encoding principles developed by Paul Lauterbur on 1972, which in turn served as a foundation for the current MRI systems. The potential of MRI for flow-related imaging is due to its coherent imaging nature. As a coherent imaging system, MRI produces complex-valued signal where magnitude and phase images can be derived. The general form of complex-valued image can be written as: I (x , y ) = I (x , y ) exp j f (x , y )
(1)
where j = -1 is the imaginary number. The magnitude part of equation (1) I (x , y ) corresponds to the magnitude image, while the phase part f (x , y ) corresponds to the phase image.
The magnitude and phase images of MRI have its own physical and clinical interpretation. MRI magnitude image represents the structural information of the object, while the phase image is related to the internal physical and chemical characteristics. Among the simultaneous use of the MRI magnitude and phase images in clinical applications is the Magnetic Resonance Angiography (MRA) procedure, first proposed by Charles Dumoulin in 1987. This procedure allows imaging and measurement of bulk blood flow in large vessels without the need for contrast agents. Quantitative flow measurement with MRI is based on the observation of phase shifts acquired by magnetic moments (spins) moving along a magnetic field gradient in comparison to stationary spins. For linear field gradients, the amount of this phase shift is proportional to the velocity of the moving spin. Phase shifts in MRI systems however not only occur from movement of magnetic moments. The originally stationary tissue may also experience phase shift come from chemical shift or external magnetic field inhomogeneity. To measure the net phase shift between moving and stationary spins, NMR signals observation is repeated in a bipolar gradient mode. Measurement in bipolar gradient mode will eliminate the phase shift accounted by another cause than movement of magnetic moments. As a consequence of the MRI phase image formation, phase shift should lies between the ranges of . Therefore, the calculation of velocity from phase shift information requires a procedure to tune the maximum phase shift measured by the system to the peak velocity expected in the vessel (encoding velocity). Encoding velocity determination is very crucial in the PC-MRA procedure. Wrapping of the velocity information will occur if the encoding velocity is too low. In the other hand, measurement will be prone to error and fail to detect slow movement if encoding velocity is too high. Peak velocity varies under different physiological conditions, and its exact determination prior to the measurement itself is not possible. However, its value can be approximated with various preliminary
techniques, either based on MRI or another measurement modality such as real time Doppler Ultrasound (Lotz et al., 2002). The next step required in the PC-MRA procedure is the calculation of blood flow volume within the vessel in every time fraction. This calculation will require an accurate estimation of vessel boundaries, so that the cross-sectional area of the vessel of interest can be correctly determined and the flowing volume fraction can be analyzed. Segmentation technique then become of high importance in this matter. Blood vessel boundaries in magnitude images correspond to the contour which is formed by pixels with large intensity gradient, as intensity gradient in magnitude images indicate tissue changes. On the other hand, blood vessel boundaries should also be the contour that encloses as much flow as possible, as appeared in the intensity of phase images. Therefore, the determination of blood vessel boundaries is required to take both the magnitude and phase information into account. The segmentation algorithm presented in this chapter is developed based on the GVFSnake active contour model. GVF-Snake Active Contour Model is a segmentation algorithm that has found its vast application in medical image analysis due to its ability to handle delicate complicated structures, in addition to its highadaptability and ease of implementation nature. Since its first introduction by Chenyang Xu and Jerry Prince in 1998, GVF-snake has been one of the most cited works in medical image segmentation. Its utilization for segmenting sophisticated anatomical structures in two and three dimension imagery has been very popular. However, little attention has been devoted to its extension for implementation in originally complex-valued images. A snake is a curve x s x s , y s , s 0, 1 in a two dimensional image field. The snake is deformable and capable of moving through an image, seeking for minimum of energy functional defined by
E=
1 1
Esnake x (s ) ds
int
( ) (E (x (s )) + E (x (s )))ds
ext
(2)
where Eint is the internal energy of the snake due to bending, and Eext is an external energy produced by the image. Since the internal energy correspond to snakes tension and rigidity that are related to the first derivative x(s) and second derivative x(s) of the snake, we may write (2) as
E=
1 0 2
2 2 a x ' (s ) + b x '' (s ) + E x (s ) ds ext
(3) where and are the weight of snakes tension and rigidity, respectively. The external energy depends on the image field I(x,y), which is considered as a continuous function, and typically is one of the following types:
1 Eext (x , y ) = - I (x , y ) 2
(4.a)
2 Eext (x , y ) = - Gs (x , y ) * I (x , y ) 3 Eext (x , y ) = I (x , y ) 2 Eext (x , y ) = Gs (x , y ) * I (x , y )
(4.b) (4.c) (4.d)
where * denotes convolution, is the gradient operator, and G(x,y) is a two-dimensional Gaussian function whose standard deviation is . Convolution by the Gaussian function is equivalent to image blurring. The last two formulations, i.e. (4.c) and (4.d), are suitable for processing an image of line drawing or a black and white image. The snake energy functional in (3) can be minimized by solving the Euler-Lagrange equation of: a (s ) x ss - b (s ) x ssss Eext x =0 (5.a)
BaCkGround
a (s ) yss - b (s ) yssss Eext y =0 (5.b) Complex-valued image processing can be formulated in the original domain of complexvalued signal, i.e. real and imaginary pair of values, or in the domain of real-valued magnitude and modulo-2 phase intensity signals, i.e. magnitude and phase intensity domain. The Complex-valued Markov Random Field (CMRF) phase unwrapping algorithm developed in Suksmono and Hirose (2002) and the snake formulation for object detection in phase images given in Suksmono et al. (2006) treat the complex-valued information as a compound magnitude and modulo-2 phase intensity images. However, in most complex-valued imaging modalities, the observable information is believed to reside in the magnitude and phase intensity representation instead of in the raw real and imaginary signal pair. Therefore, complex-valued image ssegmentation is usually performed on the magnitude and phase intensity images. This is the common pattern used in most of MRI chemical-shift and flow-related imaging applications, where MRI magnitude image is related to object structure and MRI phase image is related to the internal chemical and physical characteristics of the object. The segmentation procedure usually treats these magnitude and phase intensity images as two separated source of information. The aim of elaborating the two sources of information is to obtain accurate structural localization of the chemical or physical quantity of interest. Segmentation is then performed based on constraints that are derived from both magnitude and phase image. The previous works dedicated in PC-MRA are conducted in similar approach. Among the most prominent works in this field are those of Solemn et al. (2004) and Chung et al. (2002). Solemn et al. (2004) uses the constraint derived from the magnitude intensity gradient and the velocity-based term inferred from the phase intensity image to perform segmentation. The method shows higher numerical stability and an increasing tendency of segmentation
where xss and xssss are the second and fourth derivative of the snake along the x-axis and yss and yssss are the second and fourth derivative of the snake along the y-axis. In some cases, it is necessary to initiate objects boundary search within a large area. Usually, one may increase of the Gaussian function so that the boundary can be sensed from a large distance. However, this procedure is not always effective (Xu & Prince, 1997, 1998). A new formulation of energy minimization called gradient vector flow (GVF) field is defined as:
e
m (u
2 x
2 2 2 + uy + vx + vy + f
v - f dxdy
(6) which, by using calculus of variations, can be solved by the following Euler equations: m2u - (u - fx ) fx2 + fy2 = 0 m2u - (u - fy ) fx2 + fy2 = 0
(7.a) (7.b)
In equation (7) 2 is the Laplacian operator. Those equations can be solved further by treating u and v as functions of time and solving:
ut ( x, y, t ) =
2u ( x, y, t ) ( u ( x, y, t ) f x ( x, y ) ) f x ( x, y ) + f y ( x, y )
2
(8.a)
vt ( x, y, t ) =
2 v ( x, y, t ) ( v ( x, y, t ) f y ( x, y ) ) f x ( x, y ) + f y ( x, y )
2
(8.b) These equations are called the generalized diffusion equation where the steady-state solution (as t) converges to the Euler equation.
performance in areas of high velocity noise, due to the inclusion of magnitude intensity term. On the other hand, in cases where the magnitude representation of blood vessel is practically equal to the surrounding pixels, the velocity based term still makes the segmentation become possible. However, magnitude intensity gradient value may not yield an accurate structural constraint due to the existence of stronger non-vessel object boundaries. This is the main problem faced on the simultaneous use of magnitude and phase intensity information. Chung et al. (2002) gives the definition of speed image, i.e. the compound image of magnitude and phase information. It is formed by multiplying the magnitude image with the difference image of two phase intensity images taken in a subsequent pair of bipolar gradient. Segmentation is then performed in this compound image domain. High intensity stationary object will become less prominent in the speed image, while low intensity moving object will appear clearer. To obtain noise-free phase information, preprocessing by phase denoising filter is suggested prior to the formation of speed image. Phase unwrapping step may also be considered necessary in case the encoding velocity used in the measurement procedure were too low (Lotz et al., 2002). The speed image provides a better input formulation for image segmentation with constraints derived from both magnitude and phase intensity information. When dealing with snake active contour model, the problems which are most likely to be found are sensitivity to contour initialization, spurious edges or existence of stronger non-targeted boundaries, and local minima of energy functional. In cases such as PC-MRA blood vessel segmentation where the targeted object tends to reside in a relatively constant coordinates for every image in the data set, the problem can be solved by determining an optimum contour initialization through a series of experiment. The matter of spurious edges or stronger misleading boundaries in the other hand is trickier. The existence of such edges is most likely to prevent the snake from converging to the targeted contour, as these edges give lower
value of snake energy. Therefore, the targeted contour configuration becomes a local, instead of global minimum of snake energy functional as expected. Even when the targeted contour configuration does correspond to the global minimum of the energy functional, there is still a chance that snake does not converge to this configuration due to the downhill nature of the gradientdescent snake energy minimization scheme. This scheme searches the minimum value of a function iteratively according to the direction which gives the lowest descent of the function gradient. It guarantees the next solution to be more of a minimum than the previous one, but fails to find the global minimum of the function once it has reached and stuck in a local minimum. This characteristic also accounts for the snake sensitivity to initialization, since the initial point where the gradient descent numerical method starts to evaluate the function to be minimized may greatly affect whether the scheme would converge into a desired minimum value.
ProPosed MetHod
The method proposed in this chapter addresses two main problems found in the implementation of conventional GVF Snake in the segmentation of blood vessel from a complex-valued image. The first problem is the local minimum energy value, and the second problem is the downhill characteristics of the gradient descent numerical method adopted in the GVF Snake iteration. In the case where the targeted object boundaries were formed by the pixels with the highest intensity gradient in the whole image, GVF Snake has given satisfactory performance. However, in blood vessel segmentation from complex-valued MRI image, the vessel itself is not the most prominent object existed in the image. There are objects with higher intensity gradient in its boundaries which naturally correspond to stronger minima of the energy function. Since GVF-snake tends to converge to a stronger minimum, segmentation in such condition is likely to fail. We propose the ad-
dition of geometric shape constraint into the GVF-snake energy function, in order to make the targeted contour configuration become a stronger minimum. This additional shape constraint forces the contour to converge not only to the location of strong edges but also to a geometrical configuration that suits the characteristics of the targeted segmentation object. We named the approach as the Model-constrained GVF Snake (MC-GVF Snake). Conventional snake algorithm utilizes gradient-descent numerical method to find the minimum value of its energy function. The problem with this numerical method is that it stops its searching once a local minimum is reached. In many cases, however, the targeted contour configuration may not correspond to the first minimum found in the search. To overcome this problem, we attempted to use stochastic method in implementing the energy minimization procedure. Stochastic-snake uses random sampling with simulated annealing to estimate the minimum energy value related with targeted object boundaries. The stochastic nature gives this scheme a hill-climbing capability which makes it possible for the procedure to find stronger and more suitable minima once it has reached a local minimum.
constraint formulation is given as an additional constraint which urges the snake to preserve a circular shape during its deformation. For every snake contour which varies to time v(s,t), we define a closest circle template vc(s,t) which has its centre in the centroid of the contour and its radius as the square root of the area enclosed by the contour after divided by . Figure 1 gives an illustration of the relationship between snake contour and its corresponding circle template. The model constraint external force Fc(v(s,t)) will direct the snake control points into their closest circle template points, as given by the following mathematical formulation:
FC v (s, t ) = wc v (s, t )
arg min dist v s, t , v s, t ( ) C ( )
( (
))
(9) In this formulation, wc(v(s,t)) denotes the weighting parameter for the additional external force Fc, while [arg min(dist(v(s,t), vc(s,t)))] denotes the closest distance between each snake control points and the corresponding circle template. The GVF-snake energy minimization scheme, then become as follows:
a (s ) x ss - b (s ) x ssss + k (s ) wGVF (u ) + wC (FCx ) = 0
Model-Constrained GVf-snake
In many natural images, target object is correlated to a local, instead of global, energy minimum. In order to produce an accurate segmentation result, a snake energy function modification is required to make the target object boundary become a global minimum. One approach to the problem is to include a model constraint into the energy function formulation. The constraint should characterize a distinct feature of the target object. Furthermore, it is necessary for the constraint not to be susceptible to noise. In the case of blood vessel segmentation, due to the shape-similarity nature of the blood vessel transversal projections along the body medial axis, we propose the incorporation of a shapepreserving model constraint into the snake energy minimization scheme. The model shape
(10.a)
a (s ) yss - b (s ) yssss + k (s ) wGVF (v ) + wC (FCy ) = 0
(10.b) where wGVF and wC are the GVF and modelconstraint force weighting parameters, respectively given by: wGVF v (s, t ) = e
- arg min(dist (v (s,t ),vC (s,t )))
(11)
wC v (s, t ) = 1 - wGVF v (s, t )
(12)
Figure 1. Illustration of circle template for shape preservation
and u and v are the GVF force vector component in x and y direction, while FCx and FCy are the model constraint force vector component also in x and y direction. The weighting scheme of the external forces provides a dynamic ratio between GVF force and model-constraint force. Model constraint dominates the external force term when snake contour is far from a circular shape, while GVF force dominates the external force term when snake has already resembled a circle.
stochastic snake
Stochastic-snake utilizes simulated annealing stochastic procedure to estimate minimum energy value related to targeted object contour. This scheme is based on MetropolisHastings acceptance criteria in Markov Chain Monte-Carlo random sampling scheme. In this case, snake energy from various contour configurations is considered as an unknown distribution defined in solution space, from which Markovian random samples with the
distribution of P(x) are to be taken. Utilizing the Metropolis-Hastings criterion, the distribution of these random samples is expected to approach the actual snake energy distribution as more samples are taken, such that the minimum value of the actual snake energy distribution could be estimated from the minimum value of the random samples distribution. As mentioned earlier, the stochastic nature of this procedure gives a hill-climbing ability which provides chances to find stronger minima value as the search proceeds. However, it is the stochastic nature also which cause the procedure not to be able to guarantee the convergence to global minimum value. In a simulated annealing procedure, the Markov chain of the actual distribution of interest is modeled as a Gibbs distribution of: PT (x ) = 1 e ZT
E (x ) T
(13)
where the partition function ZT is given by:
ZT =
"x W
E (x ) T
(14)
aT = min 1, exp - DE (x *, x (t )
( {(
)})
1/T
Q(x (t ) ; x *) Q(x *; x (t ) )
In both equation, T is the annealing temperature and E is the energy functional to be minimized, or in this case the snake energy functional. At high temperature, the probability of any samples to be accepted in the Markov chain is equally likely. As the annealing temperature decreases, the distribution of Markovian random samples will converge to a uniform distribution over the global minimum of E. The probability of making a move from one random sample x(t) to another candidate sample x* is given by a sampling distribution Q(x). In this research, the sampling distribution is a Gaussian probability density function. The sampled snake contour configuration in the other hand is represented by two parameters, i.e. the coordinate of the reference point from which the radial distance of snake nodes in various angular orientations is measured and the set of radial distance of each available snake nodes with respect to the reference point. In this research, three sampling scenarios have been developed. The first scenario samples the coordinate of the reference point, assuming a constant uniform radial distance. The second scenario samples the set of radial distance, assuming a constant reference point. The last scenario samples both the reference point and the set of radial distance. The criterion of accepting or rejecting a random sample in the simulated annealing procedure is based on the Metropolis-Hastings criterion:
1/T P * (x *) Q(x (t ) ; x *) aT = min 1, (t ) (t ) P * (x ) Q(x *; x )
(16) where E is the difference of energy function value between candidate sample x* and the last accepted random sample x(t). For a complete, more detailed explanation regarding the simulated annealing procedure, readers are suggested to refer to Mc.Kay (2003). Initial temperature, cooling schedule, and final temperature determination are of important role in a simulated annealing scheme. The determination of the three parameters is given in the next paragraphs: Initial temperature is expected to provide a vast hill-climbing opportunity. In this research, initial temperature is designed to enable the maximum acceptance probability of p for random samples with considerably maximum hill-climbing energy value difference. Metropolis-Hastings criterion for the acceptance for the correspondent hill-climbing samples can be written as follows: exp -DE / T (init ) p DE
(17)
T (init ) > -
ln (p )
(18)
Cooling schedule is designed such that the acceptance probability of hill-climbing samples decreases with the factor of 1/r as the simulated annealing temperature decreases. The criterion can be expressed as follows: (t ) DE HC T (t +1) = DE (t ) + T (t ) + ln r () HC (t ) T
(15)
For the definition of P(x) given in (13), the Metropolis-Hastings criterion can be rewritten as:
(19)
where HC is the average energy value of hill-climbing samples accepted in temperature T(t).
(t)
Final temperature is defined as a temperature that gives temperature decrement gradient of less than a particular threshold value of thresh. Simulated annealing final temperature therefore can be given as follows: T ( fin -1) - T ( fin ) T ( fin -1) < thresh (20)
eXPeriMent
Algorithms are benchmarked on synthetic and MRI test image in both noise-free and noisy simulated condition. The simulated noise is additive Gaussian with mean value at 0 and standard deviation at 10% of image pixel intensity range value. Evaluation parameters consist of the required snake iteration number of snake iteration required to reach convergence and the Pratt figure of merit (FOM) (Jain, 1998) which measures the nearness of segmentation contour and referenced contour (ground truth). This last parameter is only measured for synthetic image. Figure 2 shows the example of images used in experiment. High intensity circle resides in the center of image (c) and (d) is the blood vessel cross-sectional profile which is determined as the segmentation target. Figure 3 and 4 summarize the experiment result for Model-constrained GVF Snake and the three scenarios of Stochastic Snakes. The performance of the two algorithms are compared to that of the conventional GVF Snake.
as observed from the insignificant difference in the Pratt Figure of Merit. However, on noisy condition, MC-GVF snake gives significantly better performance when compared to the GVF-snake. Stochastic-snake on the other hand is capable of producing contour with more minimum energy compared to the deterministic snake algorithms, although this is compensated with an increasing number of required computations. However, stochastic nature of the algorithm makes it totally insensitive to initialization and relatively robust to the influence of noise. Finally, in real-world implementation, both of the presented algorithms can be implemented subsequently to achieve better segmentation result. Stochastic-snake could be implemented first to create a fine initial contour for the detailed object segmentation performed using the MC-GVF snake algorithm.
future researCH direCtion

The image processing approach presented in this chapter is performed in the magnitude and phase intensity domain of the complex-valued image, instead of the original real and imaginary pair of complex-valued signals. Lottfeld et al. (2007) proposes that the consistent representation of complex-valued image is formed by the pair of real and imaginary value. Therefore, the image processing methods performed on complex-valued image should treat the image as a field of complex numbers with real and imaginary values rather than as a separated magnitude and phase intensity images. This is especially true in the case of phase unwrapping, since absolute phase information can not be extracted from the naturally wrapped representation of phase intensity images. The future research then will be directed to build a solid formulation of complex-valued image segmentation in complex domain.
ConClusion
Experiment result shows that MC-GVF snake gives faster segmentation convergence rate, indicated by fewer iteration numbers. On synthetic image where segmentation target object corresponds to the strongest edges in the image, the performance of MC-GVF snake only provide slight improvement to that of the GVF-snake,
Figure 2. (a) Magnitude image (b) Phase intensity image (c) Speed image (d) Synthetic image
Figure 3. Comparison between GVF snake and MC-GVF snake segmentation result (Green: initial contour, red: final segmentation result)
Figure 4. Stochastic snake segmentation results
referenCes
Chung, A. C. S., Noble, J. A., & Summers, P. (2002). Fusing Speed and Phase Information for Vascular Segmentation of Phase-contrast MR Angiograms. Medical Image Analysis, 6, 109128. doi:10.1016/ S1361-8415(02)00057-9 Jain, A. K. (1989). Fundamentals of Digital Image Processing. Upper Saddle River, NJ: Prentice Hall. Loffeld, O. (2008). Phase Unwrapping for SAR Interferometry A Data Fusion Approach by Kalman Filtering. IEEE Transactions on Geoscience and Remote Sensing, 46(1), 4758. doi:10.1109/ TGRS.2007.909081 Lotz, J., Meier, C., Leppert, A., & Galanski, M. (2002). Cardiovascular Flow Measurement with Phase-Contrast MR Imaging: Basic Facts and Implementation. Radiographics, 22(3), 651671. McKay, D. J. C. (2003). Information Theory, Learning, Inference, and Algorithm. Cambridge, UK: Cambridge University Press.
Solem, J. E., Persson, M., & Heyden, A. (2004). Velocity based Segmentation in Phase Contrast MRI Images. In Proceedings of MICCAI 2004, Saint-Malo, France (LNCS 3216, pp. 459-466). Suksmono, A. B., Handayani, A., & Hirose, A. (2006). Snake in Phase Domain: A Method for Boundary Detection of Objects in Phase Images. Paper presented at the World Congress in Computational Intelligence (WCCI 2006), Vancouver, BC, Canada. Suksmono, A. B., & Hirose, A. (2002). Adaptive Noise Reduction of InSAR image based on ComplexMRF Model and its Application to Phase Unwrapping Problem. IEEE Transactions on Geoscience and Remote Sensing, 40(3), 699709. doi:10.1109/ TGRS.2002.1000329 Xu, C., & Prince, J. L. (1997). Gradient Vector Flow: A New External Force for Snakes. In Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR 1997) (pp. 66-71). Xu, C., & Prince, J. L. (1998). Snakes, Shapes, and Gradient Vector Flow. IEEE Transactions on Image Processing, 7, 359369. doi:10.1109/83.661186
Astri Handayani was born in Bandung, Indonesia, in 1982. She received the bachelor and master degree in electrical engineering from the Bandung Institute of Technology, Indonesia, in 2004 and 2006, respectively. She is currently a research assistant at the Biomedical Engineering Research Division, School of Electrical Engineering and Informatics, Bandung Institute of Technology. Her research interests include medical image processing and analysis. Andriyan Bayu Suksmono graduated with BS in physics, MS in electrical engineering from Institut Teknologi Bandung, and PhD in engineering from the University of Tokyo Japan, in 1985, 1996, and 2002 respectively. He joined the Department of Electrical Engineering, Institut Teknologi Bandung in 1996 as an instructor. He became an associate professor in the School of Electrical Engineering and Informatics, ITB, since 2005. He was on leave to NIME-Japan in 1998 with JSPS Fellowship grant scheme and to University of Tokyo in 2009 through Hitachi Research Fellowship. His main research interests are Signal processing, imaging, neural networks, and applied compressive sensing. Dr. Suksmono is a member of the IEEE. Tati Mengko was born in Surabaya, Indonesia, in 1953. She received the bachelors degree in electrical engineering from the Bandung Institute of Technology, Indonesia, in 1977, and the PhD degree in image processing from the ENSERG, Institut National Polytechnique de Grenoble, France, in 1985. Dr. Mengko is currently the head of the Biomedical Engineering Research Division, School of Electrical Engineering and Informatics, Bandung Institute of Technology. Her research interests include image processing, computer vision, and pattern recognition. Akira Hirose received the PhD degree in electrical engineering from the University of Tokyo, Tokyo, Japan, in 1991. In 1987, he joined the Research Center for Advanced Science and Technology (RCAST), the University of Tokyo, as a research associate, where he was engaged in research on optical communications and measurement. In 1991, he was appointed an Instructor at the RCAST, and started neural network research. From 1993 to 1995, on leave of absence from the University of Tokyo, he was with the Institute for Neuroinformatics, University of Bonn, Bonn, Germany. He became an associate professor at the RCAST in 1995, and a professor at the Department of Electronic Engineering, The University of Tokyo, in 2007. The main fields of his current research interests are wireless electronics and neural networks. Dr. Hirose is a member of the IEEE and the JNNS.

Blood Vessel Segmentation in Complex-Valued Magnetic Resonance Images With Snake Active Contour Model

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Blood Vessel Segmentation in Complex-Valued Magnetic Resonance Images With Snake Active Contour Model

Uploaded by

Copyright:

Available Formats

International Journal of E-Health and Medical Communications, 1(1), 41-52, January-March 2010 41

2 2 a x ' (s ) + b x '' (s ) + E x (s ) ds ext

2 Eext (x , y ) = - Gs (x , y ) * I (x , y ) 3 Eext (x , y ) = I (x , y ) 2 Eext (x , y ) = Gs (x , y ) * I (x , y )

(4.b) (4.c) (4.d)

arg min dist v s, t , v s, t ( ) C ( )

- arg min(dist (v (s,t ),vC (s,t )))

wC v (s, t ) = 1 - wGVF v (s, t )

Figure 1. Illustration of circle template for shape preservation

where the partition function ZT is given by:

future researCH direCtion

Figure 4. Stochastic snake segmentation results

You might also like