PDF Regularized Image Reconstruction in Parallel Mri With Matlab 1St Edition Joseph Suresh Paul Ebook Full Chapter
PDF Regularized Image Reconstruction in Parallel Mri With Matlab 1St Edition Joseph Suresh Paul Ebook Full Chapter
PDF Regularized Image Reconstruction in Parallel Mri With Matlab 1St Edition Joseph Suresh Paul Ebook Full Chapter
https://textbookfull.com/product/design-optimization-using-
matlab-and-solidworks-1st-edition-suresh/
https://textbookfull.com/product/digital-image-interpolation-in-
matlab-first-edition-kok/
https://textbookfull.com/product/audio-and-speech-processing-
with-matlab-paul-r-hill/
https://textbookfull.com/product/matlab-parallel-computing-
toolbox-user-s-guide-the-mathworks/
MATLAB Image Processing Toolbox User s Guide The
Mathworks
https://textbookfull.com/product/matlab-image-processing-toolbox-
user-s-guide-the-mathworks/
https://textbookfull.com/product/management-of-banking-and-
financial-services-4th-edition-padmalatha-suresh-justin-paul/
https://textbookfull.com/product/medical-image-processing-
reconstruction-and-analysis-concepts-and-methods-second-edition-
jiri-jan-author/
https://textbookfull.com/product/image-processing-and-pattern-
recognition-based-on-parallel-shift-technology-first-edition-
bilan/
https://textbookfull.com/product/fortran-2018-with-parallel-
programming-1st-edition-subrata-ray-author/
Regularized Image
Reconstruction in Parallel
MRI with MATLAB®
Regularized Image
Reconstruction in Parallel
MRI with MATLAB®
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been made
to publish reliable data and information, but the author and publisher cannot assume responsibility for the validity of all
materials or the consequences of their use. The authors and publishers have attempted to trace the copyright holders of all
material reproduced in this publication and apologize to copyright holders if permission to publish in this form has not been
obtained. If any copyright material has not been acknowledged please write and let us know so we may rectify in any future
reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or utilized in
any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, micro-
filming, and recording, or in any information storage or retrieval system, without written permission from the publishers.
For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://www.copyright.
com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a
not-for-profit organization that provides licenses and registration for a variety of users. For organizations that have been granted a
photocopy license by the CCC, a separate system of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for
identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
Preface ..............................................................................................................................................xi
Acknowledgements .................................................................................................................... xiii
Authors ...........................................................................................................................................xv
v
vi Contents
xi
Acknowledgements
We are thankful to Professor Jose A Ramos, at Nova Southeastern University, for his
insightful comments on the contents of our book. We express our special gratitude to
Professor Frithjof Kruggel, Department of Biomedical Engineering, University of California,
Irvine; Professor Michael Braun, University of Technology, Sydney; and Professor Socrates
Dokos, Graduate School of Biomedical Engineering, The University of New South Wales,
Sydney, for their patience in proofreading parts of this manuscript. We also thank
Professor Mathews Jacob, University of Iowa, and Professor Albert M. Thomas, University
of California at Los Angeles, for longstanding support and collaboration. We are thankful
to Sruthi Raghoothaman, Medical Image Computing and Signal Processing Lab, Indian
Institute of Information Technology and Management–Kerala (IIITM-K), in India, for her
technical help in the theoretical development of continuation scheme. We also thank all
past and present members of the Medical Image Computing and Signal Processing Lab for
their assistance in writing this book. We thank and acknowledge the following scholars
for sharing codes for comparisons: K Dabov et al. for BM3D, E M Eksioglu for BM3D-MRI,
J Caballero et al. for DLTG, and M Lustig and J M Pauly for SPIRiT.
xiii
Authors
Raji Susan Mathew is currently pursuing her PhD in the area of magnetic resonance (MR)
image reconstruction. She received a bachelor degree in Electronics and Communication
Engineering from Mahatma Gandhi University, Kottayam, and a master’s degree in Signal
Processing from the Cochin University of Science and Technology, Kochi, in 2011 and 2013,
respectively. She is a recipient of the Maulana Azad National Fellowship (MANF) from the
University Grants Commission (UGC), India. Her research interests include regularization
techniques for MR image reconstruction and compressed sensing.
xv
1
Parallel MR Image Reconstruction
1
2 Regularized Image Reconstruction in Parallel MRI with MATLAB®
magnets. However, as fixed and resistive magnets are generally restricted to field strengths
below 0.4 Tesla (T), they cannot generate the higher field strengths necessary for high-
resolution imaging. Therefore, superconducting magnets are used in most high-resolution
imaging systems. The superconducting magnets are made from coils of superconduct-
ing wire, and they require the coils to be soaked in liquid helium to reduce their tempera-
ture to a value close to absolute zero.
1.1.4 RF Receiver
Due to the capability of RF receiver to process signals from multiple receiver coils, mod-
ern MRI systems use more than two receivers to process the signals. The bandwidth of
the received signal is dependent on the magnitude of the gradient field and is typically
less than 20 kHz. Modern receiver systems use high-input-bandwidth, high-resolution,
12–16-bit analog-to-digital converters (ADCs) with sample rates up to 100 MHz, which
eliminates the need for analog mixers in the receive chain by directly sampling the
signals.
B ( r ) = ( B0 + Gx x + Gy y + Gz z ) k = ( B0 + G ⋅ r ) k, (1.1)
Here, r = xi , yj , zk, denotes the spatial location of the spin, and the vector G = Gx Gy Gz
the field gradient, with Gx , Gy and Gz representing the gradient strengths along x, y, and z
directions, respectively. As the overall main magnetic field varies in space, the spins inside
the volume have spatially dependent resonance frequencies, expressed as:
ω ( r ) = γ ( B0 + G ⋅ r ) (1.2)
The addition of a gradient field to the static B0 field enables the resultant total magnetic
field to vary linearly in space. The Larmor frequencies also vary with spatial location as
they are proportional to the field.
Parallel MR Image Reconstruction 3
These steps for slice selection are illustrated in Figure 1.1. There are certain factors which
affect the slice properties. These include RF pulse bandwidth, RF pulse frequency and
gradient strength. The RF pulse bandwidth is the range of frequencies within the pulse.
Large bandwidth results in the selection of a larger slice, and vice versa. Varying the RF
pulse frequency moves the selected slice up or down the z-axis. The size of the slice is
determined by altering the gradient strength.
FIGURE 1.1
Slice selection and selective excitation.
4 Regularized Image Reconstruction in Parallel MRI with MATLAB®
(
Mz ( t ) = Mz 0 1 − e −t/T1 . ) (1.3)
Due to the spin-spin interactions and variations in B0 , the spins eventually lose their
transverse magnetization. The time that characterizes their return to the equilibrium state
is called spin-spin relaxation time, denoted by T2 [1]. With Mxy0 denoting the transverse
magnetization at equilibrium, the transverse magnetization Mxy is obtained as follows:
The changes in magnetization induce an alternating current in the receiver coils, and this
signal is referred to as free induction decay (FID). FID is a short-lived sinusoidal electro-
magnetic signal generated immediately following the 90° RF pulse. This signal is induced
in the receiver coil by the rotating component of the magnetization vector in the xy plane,
which crosses the coil loops perpendicularly. The FID is one among the four basic types of
NMR signals generated in different ways. The other types include the gradient echo (GRE),
induced using one RF pulse and gradient reversal; the spin echo (SE), using two RF pulses;
and the stimulated echo generated using the application of three or more RF pulses.
1.1.8 Imaging
In one-dimensional (1D) imaging, a frequency-encoding (FE) gradient pulse is applied
immediately after application of the 90° - RF pulse, as shown in Figure 1.2. The resulting
FID signal is simultaneously digitized along with application of a readout pulse. The read-
out is enabled synchronously with the FE gradient pulse. Application of this gradient
along the x-direction, modulates the precession frequency of spins as a linear function of
their positions along the x-axis. This causes the spin frequency to vary spatially:
ω ( x ) = γ ( xGx + B0 ) . (1.5)
FIGURE 1.2
Application of gradient in 1D imaging. Following the excitation with a 90° RF pulse, the FID signal is read off
in the presence of the FE gradient.
Parallel MR Image Reconstruction 5
The net signal generated from an ensemble of spins can be modeled as the summation
of their contributions from each location along the x-direction. In the case where the
spin density is modeled as a continuous function of space, the resultant signal can be
expressed as:
∞
s ( t ) = kM0 e − jγ B0t
∫
−∞
ρ ( x ) e −t/T2 ( x )e − jγ xGxt dx (1.6)
The integral in Equation (1.7) represents the Fourier transform of the spin density func-
tion. The acquired MR data can be thought of as being defined in the spatial frequency
domain (Fourier space). It follows that the spin density can be obtained by taking the
inverse Fourier transform of the signal as:
ρ ( x ) = F −1 {R ( kx )} . (1.9)
For two-dimensional (2D) image acquisition, the pulse sequence must involve two gradi-
ents. Suppose a 90° pulse excites spins within a transverse slice at a particular value of z.
In order to acquire a spatially differentiated signal from the slice, gradients in the x and y
directions are to be applied as shown in Figure 1.3. In contrast to 1D imaging, a Gy gradient
is additionally applied for a fixed time τ , before the start of the acquisition. Then the spins
along y are dephased by
θ = γ yGyτ . (1.10)
After switching off the gradient, the acquisition begins and the frequency differentiation
along the y axis is also turned off. Alternatively, the differentiation in the phase is acquired
over a short duration of the Gy pulse. Therefore, the Gy gradient is referred to as a phase-
encoding (PE) gradient.
The FID is sampled in the presence of Gx, after turning off the y gradient. The spatial
frequencies denoted by k x and k y can be defined as follows:
γ Gx t
kx =
2π
(1.11)
γ Gyτ
ky = .
2π
6 Regularized Image Reconstruction in Parallel MRI with MATLAB®
FIGURE 1.3
Application of gradients in 2D imaging. Following excitation with a 90° pulse, a phase-encoding gradient Gy
is applied in the y direction for a time τ . At the end of that time, the phase acquired by parts of the sample
along the y-axis will vary in proportion to the value of y. Then, the FID signal is read off in the presence of the
frequency-encoding gradient Gx .
For one FID signal, there is only one value of the vertical spatial frequency k y . The sequence
must be repeated several times to obtain other values of k y , with the values of Gy stepped
up or down with each onset of the FID signal generation. Once the required set of data is
obtained, the data can be arranged to form a 2D array. The signal equation in the 2D case
can then be expressed as:
∞ ∞
− j 2π ( k x x + k y y )
R ( kx , ky ) =
∫ ∫
−∞ −∞
ρ ( x, y ) e dxdy , (1.12)
where ρ varies in the xy-plane. The inverse 2D Fourier transform of the array in (1.12)
reconstructs the spin density in the image space as follows:
{
ρ ( x , y ) = F −1 R ( kx , k y ) . } (1.13)
The sampled signal is arranged along horizontal lines in the (k x k y )-plane, as shown in
Figure 1.4. The horizontal line with k y = 0 represents the FID signal sampled with Gy = 0.
For a gradient step ∆Gy , the separation between the sampled horizontal lines in the
(k x k y )-space is given by
γ∆Gyτ
∆k y = . (1.14)
2π
FIGURE 1.4
The trajectory in the k x k y plane resulting from the application of the Fourier technique of Figure 1.3. Each line
corresponds to an increment of Gy. The crosses denote the sample points.
Parallel MR Image Reconstruction 7
Until now, the discussion was limited to the acquisition in the first quadrant of the
(k x k y )-plane. Similarly, one could sample the fourth quadrant and complete the posi-
tive half-plane. This is achieved by allowing the Gy gradient to take on negative values.
The negative half-plane is acquired by allowing Gx to take negative values. Because each
step change in the gradient requires a new FID signal and the associated waiting time
TR, sampling all the quadrants becomes a lengthy process. In practice, when the gradient
pulses are rectangular, the spatial frequencies are obtained by integration over the gradi-
ent pulse duration. For time varying gradient pulses,
γ t
kx =
2π ∫ G (t ’) dt ’ ,
0
x (1.15)
and
τ
γ
ky =
2π ∫ G (τ ’) dτ ’.
0
y
FIGURE 1.5
k-space sampling and Nyquist criterion. The k-space sampling interval ∆k x and maximum extent k xmax are related
to the FOVx and voxel size ∆x, respectively, of the reconstructed image in x-direction. Similarly, ∆k y and k ymax are
related to the FOVy and ∆y.
is band limited, and the corresponding full object in the image space. The band limits
in the x and y dimensions are FOVx and FOVy , respectively. The central panel shows the
sampling grid on the k-space, and the corresponding object space with FOVx = 2Wx and
FOVy = 2Wy . As in the 1D case, the object space is repeated periodically with a repetition
interval equal to the FOV. Here, the object images do not overlap, and hence the object of
interest can be recovered by application of a low-pass filter. The rightmost panel illustrates
the case in which FOVx < 2Wx and FOVy < 2Wy . In this case, the object images overlap, and
a perfect recovery of the object is not possible.
FOV depends on the strengths of magnetic gradients, their durations and the acquisi-
tion time. By acquiring data at uniform intervals ∆ts in time, the smallest FE step in the
k-space will be ∆k x = γ Gx ∆ts /2π . This in turn determines the size of FOV. FOV should be
set appropriately to cover the whole imaged object. Since the encoding function is peri-
odic, parts of the imaged object outside the FOV are misinterpreted by aliasing as being
inside the FOV. While designing an MRI acquisition, appropriate choices of the FOV and
the spatial resolution (i.e., voxel size) are to be determined. As the magnitude of the gra-
dient fields cannot be increased beyond a limit, the total acquisition time is proportional
to the number and length of the scan lines acquired, in the case of Cartesian imaging.
In this case, a larger FOV (or smaller voxels) in the PE direction requires more scan lines.
However, acquisition time is unaffected by the spacing between samples within a scan
line, and arbitrarily large FOV in the FE direction can be achieved for free. Therefore, the
longest dimension of the volume can be chosen as FE direction so that one can overs-
ample the k-space in that direction to avoid any possibility of aliasing.
Parallel MR Image Reconstruction 9
FIGURE 1.6
The steps in a typical SE pulse sequence. Corresponding to each step, the k-space trajectories and location are
shown in the right side.
Here data is sampled in a line-by-line fashion on a rectangular grid. The main benefit
of this type of sampling trajectory is that images can be directly reconstructed using fast
Fourier transform (FFT). The spacing and extent of the Cartesian grid have a direct relation
to the acquisition time, FOV and voxel size of the resulting image. In Cartesian acquisi-
tion, the number of PE steps can be reduced by an acceleration factor or reduction factor
R by increasing the distance between the k-space lines. If only one line is left between
two k-space lines, the MR data acquisition is said to be accelerated by a factor of R = 2.
The reduction of sampling along the PE direction of the k-space results in a reduced FOV
in that direction with associated fold-over artifacts.
Parallel MR Image Reconstruction 11
FIGURE 1.7
Cartesian k-space acquisition. (a): Linear phase-encoding order or standard MR acquisition, (b): low-high or
centric k-space order.
The position of the line being filled in the k y direction is determined by the PE gradient.
The amplitude of the PE gradient is incremented in steps such that the next adjacent line
in the k-space is filled successively starting from one edge of the k-space until the opposite
edge is reached. This is illustrated in Figure 1.7 and is referred to as a linear phase-encoding
order [2]. In some of the dynamic applications, including contrast enhanced angiography,
a different ordering of the PE steps is adopted to retain the contrast information at the
onset of acquisition. In that case, the PE gradient is incremented from zero, but with an
alternating sign, starting from the centre of the k-space in steps to acquire lines in extreme
edges. This type of k-space sampling in the PE direction is referred to as the centric phase
encode order or low-high k-space order.
FIGURE 1.8
k-space and image are Fourier transform (FT) pairs. Panels in the top row show a completely filled k-space and
the resulting image. The middle and bottom rows illustrate the k-space and image obtained after truncating the
k-spaces with large and small rectangular windows. The effects of truncation are clearly seen in the zoomed
versions of the images shown in the rightmost panels.
As inferred from Figure 1.8, with a greater net strength of the phase-encoding gradient,
the data is mapped to a farther position from the k-space centre. Based on the polarity
of the gradient pulse, the mapping can be either toward the upper or lower directions.
Thus, the duration of the PE gradient governs the location of data on the vertical axis of the
k-space. Alternatively, with a greater net strength of the FE gradient, the data samples tend
to be located farther from the k-space centre. The data is mapped to the right side of the
k-space if the gradient is positive, or in the left direction if the gradient is negative. Thus,
the strength of the FE gradient determines the location of data on the horizontal axis of
k-space. After the acquisition of the entire k-space, inverse Fourier transform can be used
to reconstruct the image. With uniform Cartesian acquisition, the k-space data is filled dur-
ing the scan so that one phase-encoding line is collected per TR. An extensive coverage of
the effect of discrete k-space sampling are provided in [1,4].
Scanning time increases with an increase in the number of PE lines. In the case of 3D
imaging, second and third dimensions are acquired using phase-encoding. Compared to
frequency-encoding, phase-encoding is a much slower process. As an example, with a
TR of 500 ms, collection of 256 PE lines require 128 seconds in comparison to 8 ms for
acquisition of 256 samples in one PE line [5]. This indicates that by reducing the number of
acquired PE lines, the data acquisition time can be significantly reduced. The image recon-
structed from full k-space can achieve the best quality as all signals have been sampled
to fill the entire k-space. Considering the scanning time required, k-space reconstruction
is carried out from under-sampled k-space, in which only a limited number of k-space
samples are acquired.
Another random document with
no related content on Scribd:
The Project Gutenberg eBook of
Henkivakuutusherroja
This ebook is for the use of anyone anywhere in the United States
and most other parts of the world at no cost and with almost no
restrictions whatsoever. You may copy it, give it away or re-use it
under the terms of the Project Gutenberg License included with this
ebook or online at www.gutenberg.org. If you are not located in the
United States, you will have to check the laws of the country where
you are located before using this eBook.
Title: Henkivakuutusherroja
Romaani
Language: Finnish
Romaani
Kirj.
VEIKKO KORHONEN
— Ai perhana!
— Ka, sinä lempoko siinä. Kun kulkee kuin lehmä kadulla, alkoi
Varsala purkaa sisuaan.
— Älä.
— Vai pääsi poika ylenemään. No, nythän sinä saat tuntea, miten
ihanaa on olla henkivakuutusherrana. Useimmiten toiset tekevät
työn, ja me saamme niittää sadon, ainakin mitä tantiemiin tulee.
Meidän on tosin pidettävä huoli kilpailusta ja siitä, että asiamiehet,
joille varsinainen työnteko kuuluu, eivät pääse laiskottelemaan,
mutta sehän tehtävä onkin kaikkein helpoimpia. Ja ajatteles, ethän
sinä enemmän kuin minäkään, tullessani Leimausyhtiön
palvelukseen ole suorittanut mitään tutkintoja tällä alalla, ainoastaan
ovelasti päättänyt muutamia tukkikauppoja maalaisten kanssa. Niistä
minut napattiin tähän toimeen. Huomasivat kai, että minussa on
liikeneroa ja että osaan olla ovela. Ja hyvin minä olen
menestynytkin, osaan herättää ihmisissä, varsinkin maalaisissa
luottamusta. Sen vaikuttaa ulkoasuni. Minä näes en komeile
vaatteilla. Olen sattunut usein toisten yhtiöitten samaan luokkaan
kuuluvien virkamiesten kanssa samaan aikaan hankinnalle ja tullut
huomaamaan, että kovin koreaa herrasmiestä katsellaan kansan
seassa hiukan sekavin tuntein. Minulla on siitä tuoreita esimerkkejä.
Viime viikolla yövyin erääseen taloon, jossa hieroin kymmenen
tuhannen vakuutusta, jota oli samana päivänä kärttänyt sinullekin
hyvin tunnettu Rientoyhtiön Keikaus. Kuulin miten isäntä arveli
emännälleen: — Tuo näyttää oikealta mieheltä! Se Keikaus oli niin
hieno mies ja isovatsainen. Taitaa syödä vakuutettaviensa varoja.
Mitäs, jos ottaisimmekin vakuutuksen Leimaus yhtiössä. Ja aamulla
se vakuutus päätettiin.
— Jospa se lienee niin kuin sanot. Mutta mitä muuta sanoisit vielä
ihmisiä piinatessa tarvittavan?
Varsala naurahti.
— Häh?
— Häh?
— Nousenhan minä.
— Ei ole.
— Kyllä.