Visit https://ebooknice.

com to download the full version and

explore more ebooks

(Ebook) Signal processing for magnetic resonance

imaging and spectroscopy by Hong Yan ISBN
9780203908785, 9780824706531, 0203908783, 0824706536

_____ Click the link below to download _____

https://ebooknice.com/product/signal-processing-for-
magnetic-resonance-imaging-and-spectroscopy-875452

Explore and download more ebooks at ebooknice.com

Here are some recommended products that might interest you.
You can download now and explore!

(Ebook) Biota Grow 2C gather 2C cook by Loucas, Jason; Viles, James

ISBN 9781459699816, 9781743365571, 9781925268492, 1459699815,
1743365578, 1925268497

https://ebooknice.com/product/biota-grow-2c-gather-2c-cook-6661374

ebooknice.com

(Ebook) Matematik 5000+ Kurs 2c Lärobok by Lena Alfredsson, Hans

Heikne, Sanna Bodemyr ISBN 9789127456600, 9127456609

https://ebooknice.com/product/matematik-5000-kurs-2c-larobok-23848312

ebooknice.com

(Ebook) SAT II Success MATH 1C and 2C 2002 (Peterson's SAT II Success)

by Peterson's ISBN 9780768906677, 0768906679

https://ebooknice.com/product/sat-ii-success-
math-1c-and-2c-2002-peterson-s-sat-ii-success-1722018

ebooknice.com

(Ebook) Hyperpolarized Carbon-13 Magnetic Resonance Imaging and

Spectroscopy by Larson, Peder; ISBN 9780128222690, 0128222697

https://ebooknice.com/product/hyperpolarized-carbon-13-magnetic-
resonance-imaging-and-spectroscopy-37036818

ebooknice.com
(Ebook) Master SAT II Math 1c and 2c 4th ed (Arco Master the SAT
Subject Test: Math Levels 1 & 2) by Arco ISBN 9780768923049,
0768923042

https://ebooknice.com/product/master-sat-ii-math-1c-and-2c-4th-ed-
arco-master-the-sat-subject-test-math-levels-1-2-2326094

ebooknice.com

(Ebook) Cambridge IGCSE and O Level History Workbook 2C - Depth Study:

the United States, 1919-41 2nd Edition by Benjamin Harrison ISBN
9781398375147, 9781398375048, 1398375144, 1398375047

https://ebooknice.com/product/cambridge-igcse-and-o-level-history-
workbook-2c-depth-study-the-united-states-1919-41-2nd-edition-53538044

ebooknice.com

(Ebook) Advanced Image Processing in Magnetic Resonance Imaging by

Luigi Landini, Vincenzo Positano, Maria Santarelli ISBN 9780824725426,
0824725425

https://ebooknice.com/product/advanced-image-processing-in-magnetic-
resonance-imaging-1098214

ebooknice.com

(Ebook) Introduction to functional magnetic resonance imaging:

Principles and techniques by Richard B. Buxton ISBN 9780521899956,
0521899958

https://ebooknice.com/product/introduction-to-functional-magnetic-
resonance-imaging-principles-and-techniques-1431252

ebooknice.com

(Ebook) Coding and signal processing for magnetic recording systems by

Bane Vasic, Erozan M. Kurtas ISBN 9780849315244, 0849315247

https://ebooknice.com/product/coding-and-signal-processing-for-
magnetic-recording-systems-1185338

ebooknice.com
Signal processing for magnetic resonance imaging and
spectroscopy 1st Edition Hong Yan Digital Instant
Download
Author(s): Hong Yan
ISBN(s): 9780824706531, 0824706536
Edition: 1
File Details: PDF, 8.61 MB
Year: 2002
Language: english
 Signal Processing for
Magnetic Resonance
Imaging and Spectroscopy

edited by
Hong Yan
University of Sydney
Sydney, Australia

Marcel Dekker, Inc. New York • Basel

TM

Copyright © 2002 by Marcel Dekker, Inc. All Rights Reserved.

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 ISBN: 0-8247-0653-6

This book is printed on acid-free paper.

Headquarters
Marcel Dekker, Inc.
270 Madison Avenue, New York, NY 10016
tel: 212-696-9000; fax: 212-685-4540

Eastern Hemisphere Distribution

Marcel Dekker AG
Hutgasse 4, Postfach 812, CH-4001 Basel, Switzerland
tel: 41-61-261-8482; fax: 41-61-261-8896

World Wide Web

http://www.dekker.com

The publisher offers discounts on this book when ordered in bulk quantities. For
more information, write to Special Sales/Professional Marketing at the headquarters
address above.

Copyright 䉷 2002 by Marcel Dekker, Inc. All Rights Reserved.

Neither this book nor any part may be reproduced or transmitted in any form or by
any means, electronic or mechanical, including photocopying, microfilming, and
recording, or by any information storage and retrieval system, without permission
in writing from the publisher.

Current printing (last digit):

10 9 8 7 6 5 4 3 2 1

PRINTED IN THE UNITED STATES OF AMERICA

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Series Introduction

Over the past 50 years, digital signal processing has evolved as a

major engineering discipline. The fields of signal processing have
grown from the origin of fast Fourier transform and digital filter
design to statistical spectral analysis and array processing, image,
audio, and multimedia processing, and shaped developments in high-
performance VLSI signal processor design. Indeed, there are few
fields that enjoy so many applications—signal processing is
everywhere in our lives.
When one uses a cellular phone, the voice is compressed, coded,
and modulated using signal processing techniques. As a cruise missile
winds along hillsides searching for the target, the signal processor is
busy processing the images taken along the way. When we are
watching a movie in HDTV, millions of audio and video data are
being sent to our homes and received with unbelievable fidelity.
When scientists compare DNA samples, fast pattern recognition
techniques are being used. On and on, one can see the impact of
signal processing in almost every engineering and scientific
discipline.
Because of the immense importance of signal processing and the
fast-growing demands of business and industry, this series on signal
processing serves to report up-to-date developments and advances in
the field. The topics of interest include but are not limited to the
following:

· Signal theory and analysis

· Statistical signal processing
· Speech and audio processing
· Image and video processing
· Multimedia signal processing and technology
· Signal processing for communications
· Signal processing architectures and VLSI design

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 vi SERIES INTRODUCTION

We hope this series will provide the interested audience with

high-quality, state-of-the-art signal processing literature through
research monographs, edited books, and rigorously written textbooks
by experts in their fields.

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Signal Processing and Communications

Editorial Board
Maurice G. Ballanger, Conservatoire National
des Arts et Métiers (CNAM), Paris
Ezio Biglieri, Politecnico di Torino, Italy
Sadaoki Furui, Tokyo Institute of Technology
Yih-Fang Huang, University of Notre Dame
Nikhil Jayant, Georgia Tech University
Aggelos K. Katsaggelos, Northwestern University
Mos Kaveh, University of Minnesota
P. K. Raja Rajasekaran, Texas Instruments
John Aasted Sorenson, IT University of Copenhagen

1. Digital Signal Processing for Multimedia Systems, edited by Keshab

K. Parhi and Takao Nishitani
2. Multimedia Systems, Standards, and Networks, edited by Atul Puri
and Tsuhan Chen
3. Embedded Multiprocessors: Scheduling and Synchronization, Sun-
dararajan Sriram and Shuvra S. Bhattacharyya
4. Signal Processing for Intelligent Sensor Systems, David C. Swanson
5. Compressed Video over Networks, edited by Ming-Ting Sun and Amy
R. Reibman
6. Modulated Coding for Intersymbol Interference Channels, Xiang-Gen
Xia
7. Digital Speech Processing, Synthesis, and Recognition: Second Edi-
tion, Revised and Expanded, Sadaoki Furui
8. Modern Digital Halftoning, Daniel L. Lau and Gonzalo R. Arce
9. Blind Equalization and Identification, Zhi Ding and Ye (Geoffrey) Li
10. Video Coding for Wireless Communication Systems, King N. Ngan,
Chi W. Yap, and Keng T. Tan
11. Adaptive Digital Filters: Second Edition, Revised and Expanded,
Maurice G. Bellanger
12. Design of Digital Video Coding Systems, Jie Chen, Ut-Va Koc, and
K. J. Ray Liu

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 13. Programmable Digital Signal Processors: Architecture, Program-
ming, and Applications, edited by Yu Hen Hu
14. Pattern Recognition and Image Preprocessing: Second Edition, Re-
vised and Expanded, Sing-Tze Bow
15. Signal Processing for Magnetic Resonance Imaging and Spectros-
copy, edited by Hong Yan
16. Satellite Communication Engineering, Michael O. Kolawole

Additional Volumes in Preparation

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Preface

Magnetic resonance imaging (MRI) and magnetic resonance spectroscopy

(MRS) have gained widespread use for medical diagnosis in recent years.
The potential also exists for magnetic resonance to be useful in building
quantum computers, which may have the capacity of being billions of times
faster than classical computers. An important step in the use of MRI and
MRS is the processing of detected signals. Typical tasks include image re-
construction from complete or incomplete k-space data, removal of artifacts,
segmentation of an image into homogeneous regions, analysis of the shape
and motion of the heart, processing and visualization of functional MR im-
ages, characterization of brain tissues, and estimation of the parameters of
a spectroscopic signal. This book provides a review of prevalent signal pro-
cessing algorithms for solving these problems and reports the latest devel-
opments in the field.
The book is divided into three parts. Part I discusses algorithms for
image reconstruction from MR signals, removal of image distortions and
artifacts, and image visualization. Chapter 1 reviews data sampling require-
ments in the k-space and provides basic mathematical formulations of image
reconstruction problems. Chapter 2 presents algorithms for the reconstruc-
tion of wavelet transform coefficients directly from the Randon transform
domain. The algorithms can be useful for image enhancement, feature ex-
traction, and the study of local tomography. Chapter 3 describes the con-
volution regridding method for image reconstruction from data obtained on
v

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 vi Preface

a nonuniform grid in the k-space. Chapter 4 develops a technique based on

a dynamic finite element mesh (DMESH) model to analyze object motion
and distortions in MRI. Chapter 5 presents a method based on the projection
onto convex sets (POCS) algorithm to reduce the artifacts caused by un-
wanted motion of an object during data acquisition. Chapter 6 provides an
overview of several tagged MRI methods for fast imaging of the heart mo-
tion. Chapter 7 presents a cortical flattening method for the visualization of
functional MRI (fMRI) data and a k-space correlation method for the anal-
ysis of the data based on both magnitude and phase information.
Part II covers techniques for the extraction of meaningful regions in
MR images, as well as the analysis of shapes, geometric features, and dy-
namic information. Chapter 8 presents an image segmentation algorithm
based on a multiscale linking model, which can provide the gross shape
information at a higher scale and subvoxel accuracy at a lower scale. Chapter
9 proposes a method for cortical surface segmentation by deforming a cel-
lular object initialized on the volume of the brain toward the interior of the
deep cortical convolutions. Chapter 10 reviews feature-based image analysis
methods including registration, segmentation, and supervised and unsuper-
vised classification. Chapter 11 presents an fMRI segmentation method that
takes voxel connectivity information into account and a method for activity
detection based on anisotropic diffusion. Chapter 12 describes image seg-
mentation techniques based on neural network models including learning
vector quantization (LVQ), the multilayer perceptron, and self-organizing
feature maps (SOFM). Chapter 13 introduces an approach to image seg-
mentation based on stochastic pixel and context models. Chapter 14 presents
a frequency analysis-based method and a local principal component analysis
(PCA) based method for brain activation signal detection in fMRI. Chapter
15 provides an overview of 2-D and 3-D image analysis methods for mea-
suring the deformation of the heart in tagged images.
Part III discusses methods for spectral estimation and analysis of MRS
signals and their applications. Chapters 16 and 17 provide an overview of
one-dimensional and multidimensional spectroscopic signal processing
methods, respectively. The linear prediction, the state-space model, and the
maximum entropy based methods described in these chapters can overcome
several disadvantages of the conventional Fourier transform-based method
and provide improved spectral quantification. Chapter 18 reviews several
spectroscopic imaging methods that can provide fast data acquisition and
improve image quality compared to the traditional Fourier transform-based
chemical shift imaging method. Chapter 19 reviews techniques for brain
tissue characterization and tumor identification by analysis of the MR spectra
based on statistical pattern recognition algorithms. Chapter 20 describes a
wavelet-pocket based algorithm for analysis of metabolic peak parameters

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Preface vii

in MR spectra. Chapter 21 presents the Cramér–Rao theory for analyzing

the lower bounds of statistical errors in parameter estimation in spectroscopic
signal processing. Error bounds can be used to determine the quality of an
input signal or the performance of a parameter estimation procedure.
Sophisticated signal processing methods can serve as useful tools for
a cost-effective and timely solution to real-world problems. This book pre-
sents a number of ready-to-use algorithms, as well as new research results
and directions in MR signal processing. Its intended audience includes re-
searchers, graduate students, and senior undergraduate students working on
MRI- and MRS in radiology, physics, chemistry, electrical engineering, and
computer science, and hardware and software engineers working on MRI-
and MRS-related products and services. Readers of this book will be able
to utilize the techniques presented to solve the problems at hand and find
inspiration and encouragement to develop new or improved methods for
MRI and MRS.
I would like to thank the authors for their contributions to this book.
I am grateful to Mrs. Inge Rogers for her clerical work and proofreading
and to several referees for providing technical comments on the chapters.
Most of the editing was done when I was on leave in Hong Kong from the
University of Sydney and support from a research grant from the Faculty of
Science and Engineering of the City University of Hong Kong was greatly
appreciated. Finally, I would like to thank the editors at Marcel Dekker, Inc.,
for their help during the writing of this book.

Hong Yan

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Contents

Series Introduction K. J. Ray Liu iii

Preface v
Contributors xiii

PART I. IMAGE RECONSTRUCTION AND

RESTORATION

1. Introduction to Image Reconstruction 1

Zhi-Pei Liang, Jim Ji, and E. Mark Haacke

2. Wavelet-Based Multiresolution Local Tomography 25

F. Rashid-Farrokhi and K. J. R. Liu

3. The Point Spread Function of Convolution Regridding

Reconstruction 59
Gordon E. Sarty

4. Mapping Motion and Strain with MRI 91

Yudong Zhu

ix

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 x Contents

5. Rotational Motion Artifact Correction Based on Fuzzy

Projection onto Convex Sets 125
Chaminda Weerasinghe, Lilian Ji, and Hong Yan

6. Tagged MR Cardiac Imaging 167

Nikolaos V. Tsekos and Amir A. Amini

7. Functional MR Image Visualization and Signal Processing

Methods 189
Alex R. Wade, Brian A. Wandell, and Thomas P. Burg

PART II. IMAGE SEGMENTATION AND ANALYSIS

8. Multiscale Segmentation of Volumetric MR Brain Images 209

Wiro J. Niessen, Koen L. Vincken, Joachim Weickert, and
Max A. Viergever

9. A Precise Segmentation of the Cerebral Cortex from 3-D

MRI Using a Cellular Model and Homotopic Deformations 239
Yann Cointepas, Isabelle Bloch, and Line Garnero

10. Feature Space Analysis of MRI 255

Hamid Soltanian-Zadeh

11. Geometric Approaches for Segmentation and Signal

Detection in Functional MRI Analysis 317
Guillermo Sapiro

12. MR Image Segmentation and Analysis Based on Neural

Networks 341
Javad Alirezaie and M. E. Jernigan

13. Stochastic Model Based Image Analysis 365

Yue Wang and Tülay Adaı

14. Functional MR Image Analysis 401

Shang-Hong Lai and Ming Fang

15. Tagged MR Image Analysis 429

Amir A. Amini and Yasheng Chen

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Contents xi

PART III. SPECTROSCOPIC SIGNAL PROCESSING

16. Time-Domain Spectroscopic Quantitation 475

Leentje Vanhamme, Sabine Van Huffel, and Paul Van Hecke

17. Multidimensional NMR Spectroscopic Signal Processing 509

Guang Zhu and Yingbo Hua

18. Advanced Methods in Spectroscopic Imaging 545

Keith A. Wear

19. Characterization of Brain Tissue from MR Spectra for

Tumor Discrimination 569
Paulo J. G. Lisboa, Wael El-Deredy, Y. Y. Barbara Lee,
Yangxin Huang, Angelica R. Corona Hernandez,
Peter Harris, and Carles Arús

20. Wavelet Packets Algorithm for Metabolite Quantification

in Magnetic Resonance Spectroscopy and Chemical
Shift Imaging 589
Luca T. Mainardi, Sergio Cerutti, Daniela Origgi, and
Giuseppe Scotti

21. Cramér-Rao Bound Analysis of Spectroscopic Signal

Processing Methods 613
Sophie Cavassila, Dirk van Ormondt, and
Danielle Graveron-Demilly

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Contributors

Tülay Adalı Ph.D. Associate Professor, Computer Science and Electri-

cal Engineering, University of Maryland Baltimore County, Baltimore,
Maryland

Javad Alirezaie, Ph.D. Assistant Professor, Electrical and Computer En-

gineering, Ryerson University, Toronto, Ontario, Canada

Amir A. Amini, Ph.D. Associate Professor and Director, Biomedical En-

gineering, Medicine, and Radiology, Washington University School of Med-
icine in St. Louis, St. Louis, Missouri

Carles Arús, Ph.D. Assistant Professor, Department of Biochemistry and

Molecular Biology, Universitat Autònoma de Barcelona, Barcelona, Spain

Isabelle Bloch, Ph.D., H.D.R. Professor, Signal and Image Processing,

Ecole Nationale Supérieure des Télécommunications, Paris, France

Thomas P. Burg, Dipl.Phys. Physicist, Electrical Engineering and Com-

puter Science, Massachusetts Institute of Technology, Cambridge, Massa-
chusetts

Sophie Cavassila, Ph.D. Assistant Professor, CNRS UMR, Université

Claude Bernard, Lyon 1, Villeurbanne, France
xiii

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 xiv Contributors

Sergio Cerutti, M.D. Professor, Department of Biomedical Engineering,

Polytechnic University, Milan, Italy

Yasheng Chen, B.Sc., M.Sc. Department of Biomedical Engineering,

Washington University School of Medicine in St. Louis, St. Louis, Missouri

Yann Cointepas, Ph.D. Signal and Image Processing, Ecole Nationale

Supérieure des Télécommunications, Paris, France

Wael El-Deredy, Ph.D. School of Computing and Mathematical Sciences,

John Moores University, Liverpool, United Kingdom

Ming Fang, Ph.D. Project Manager, Imaging and Visualization, Siemens

Corporate Research, Princeton, New Jersey

Line Garnero, Ph.D. Director of Research, Cognitive Neuroscience and

Cerebral Imaging, Hôpital La Salpêtrière, Paris, France

Danielle Graveron-Demilly, Ph.D. Professor, CNRS UMR, Université

Claude Bernard Lyon I, Villeurbanne, France

E. Mark Haacke, Ph.D. Director, The Magnetic Resonance Imaging In-

stitute for Biomedical Research, St. Louis, Missouri

Peter Harris, B.Sc., Ph.D. School of Computing and Mathematical Sci-

ences, John Moores University, Liverpool, United Kingdom

Angelica R. Corona Hernandez, B.Eng., M.Sc. School of Computing

and Mathematical Sciences, John Moores University, Liverpool, United
Kingdom

Yangxin Huang, Ph.D. Biostatistician, Frontier Science and Technology

Research, Harvard School of Public Health, Chestnut Hill, Massachusetts

Yingbo Hua, Ph.D. Professor, Department of Electrical Engineering, Uni-

versity of California, Riverside, California

M. E. Jernigan, Ph.D. Professor, Systems Design Engineering, University

of Waterloo, Waterloo, Ontario, Canada

Jim Ji Department of Electrical and Computer Engineering, University of

Illinois at Urbana-Champaign, Urbana, Illinois

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Contributors xv

Lilian Ji, M.E. School of Electrical and Information Engineering, Uni-

versity of Sydney, Sydney, Australia

Shang-Hong Lai, Ph.D. Professor, Department of Computer Science, Na-

tional Tsing-Hua University, Hsinchu City, Taiwan

Y. Y. Barbara Lee, B.Sc., Ph.D. School of Computing and Mathematical

Sciences, John Moores University, Liverpool, United Kingdom

Zhi-Pei Liang, Ph.D. Professor, Department of Electrical and Computer

Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois

Paulo J. G. Lisboa, B.Sc., Ph.D. Professor, School of Computing and

Mathematical Sciences, John Moores University, Liverpool, United
Kingdom

K. J. Ray Liu, Ph.D. Professor, Department of Electrical and Computer

Engineering, University of Maryland, College Park, Maryland

Luca T. Mainardi, Ph.D. Professor, Department of Biomedical Engineer-

ing, Polytechnic University, Milan, Italy

Wiro J. Niessen, Ph.D. Image Sciences Institute, University Medical Cen-

ter, Utrecht, The Netherlands

Daniela Origgi, Ph.D. Medical Physics, European Institute of Oncology,

Milan, Italy

F. Rashid-Farrokhi, Ph.D. Wireless Research Department, Lucent Tech-

nologies, Holmdel, New Jersey

Guillermo Sapiro, Ph.D. Associate Professor, Department of Electrical

and Computer Engineering, University of Minnesota, Minneapolis,
Minnesota

Gordon E. Sarty, Ph.D. Assistant Professor, Department of Psychology,

University of Saskatchewan, Saskatoon, Saskatchewan, Canada

Giuseppe Scotti, M.D. Professor, Department of Neuroradiology, San Raf-

faele Hospital, Milan, Italy

Hamid Soltanian-Zadeh, Ph.D. Senior Staff Scientist, Department of Ra-

diology, Henry Ford Health System, Detroit, Michigan and Associate Pro-

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 xvi Contributors

fessor, Department of Electrical and Computer Engineering, University of

Tehran, Tehran, Iran

Nikolaos V. Tsekos, Ph.D. Assistant Professor, Radiology, Washington

University School of Medicine in St. Louis, St. Louis, Missouri

Leentje Vanhamme, Ph.D. Department of Electrical Engineering, Kath-

olieke Universiteit Leuven, Leuven, Belgium

Paul Van Hecke, Ph.D. Professor, Biomedical NMR Unit, Katholieke

Universiteit Leuven, Leuven, Belgium

Sabine Van Huffel, Ph.D., M.D. Professor, Department of Electrical En-

gineering, Katholieke Universiteit Leuven, Leuven, Belgium

Dirk van Ormondt, Ph.D. Associate Professor, Applied Physics, Delft

University of Technology, Delft, The Netherlands

Max A. Viergever, Ph.D. Professor, Image Sciences Institute, University

Medical Center, Utrecht, The Netherlands

Koen L. Vincken, Ph.D. Image Sciences Institute, University Medical

Center, Utrecht, The Netherlands

Alex R. Wade, Ph.D. Department of Psychology, Stanford University,

Stanford, California

Brian A. Wandell, Ph.D. Professor, Department of Psychology, Stanford

University, Stanford, California

Yue Wang, Ph.D. Associate Professor, Electrical Engineering and Com-

puter Science, The Catholic University of America, Washington, D.C.

Keith A. Wear, Ph.D. Research Physicist, Center for Devices and Radio-
logical Health, U.S. Food and Drug Administration, Rockville, Maryland

Chaminda Weerasinghe, Ph.D. Research Engineer, School of Electrical

and Information Engineering, University of Sydney, Sydney, Australia

Joachim Weickert, Ph.D. Mathematics and Computer Science, University

of Mannheim, Mannheim, Germany

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Contributors xvii

Hong Yan, Ph.D. Professor, School of Electrical and Information Engi-

neering, University of Sydney, Sydney, Australia

Guang Zhu, Ph.D. Associate Professor, Department of Biochemistry, The

Hong Kong University of Science and Technology, Clear Water Bay, Kow-
loon, Hong Kong, China

Yudong Zhu, Ph.D. Senior Scientist, Corporate Research and Develop-

ment, General Electric, Schenectady, New York

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 1
Introduction to Image Reconstruction

Zhi-Pei Liang and Jim Ji

University of Illinois at Urbana-Champaign, Urbana, Illinois
E. Mark Haacke
Magnetic Resonance Imaging Institute for Biomedical
Research, St. Louis, Missouri

1 INTRODUCTION
1.1 Basic Imaging Equation
In MRI, the desired image function I(⭈), representing spin density distribu-
tions weighted by relaxation effects, diffusion effects, etc., can be encoded
in the measured data S(⭈) in a variety of ways [1–4]. This chapter focuses
on the popular Fourier imaging scheme1 [4,5], in which I(⭈) is related to S(⭈)
by the Fourier integral

→
S( k ) = Ᏺ{I} = 冕 → →

I( →r )e⫺i2␲ k ⭈ r d →r (1)

1
We consider back-projection imaging [1,6] as a special case of Fourier imaging. Several non-
Fourier imaging schemes exist, including the Hadamard transform method [7], the wavelet
transform based methods [8–12], and SVD (singular value decomposition) based methods
[13–16], which are not covered in this chapter.

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 2 Liang et al.

→ →
In the two-dimensional case considered here, k = (kx , ky), r = (x, y), and the
imaging equation can be written explicitly as

冕冕
⬁ ⬁

S(kx , ky) = I(x, y)e⫺i2␲(kx x⫹ky y) dx dy (2)

⫺⬁ ⫺⬁

1.2 Sampling Requirements of k-Space Signals

In practice, S(kx , ky) is collected at a discrete set of k-space points. Two
popular data collection schemes, known as rectilinear sampling and polar
sampling, are shown in Fig. 1. It is clear from the figure that

再 kx = m ⌬kx
ky = n ⌬ky
(3)

for rectilinear sampling, and

再 kx = m ⌬k cos(n ⌬␾)
ky = m ⌬k sin(n ⌬␾)
(4)

for polar sampling, where m and n take integer values. We next discuss the
Nyquist requirement on selecting (⌬kx , ⌬ky) and (⌬k, ⌬␾).
Let us first review some basic definitions and results of standard sam-
pling theory [17–19].

FIGURE 1 Two basic k-space sampling schemes used in MRI experi-

ments: (a) rectilinear sampling and (b) polar sampling.

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Introduction to Image Reconstruction 3

Definition 1 A signal g(x) is said to be support limited if there exists a

finite W such that g(x) = 0 for 兩x兩 ⱖ W, and 兩x兩 < W is called the spatial
support of g(x).
Definition 2 A signal g(x) is said to be (frequency) band limited if its
frequency spectrum G(k) = Ᏺ{g} is zero for 兩k兩 ⱖ kmax , where kmax is called
the frequency bandwidth of the signal.
Theorem 1 (Shannon) A band-limited function can be reconstructed per-
fectly from its sampled values taken uniformly at an interval not exceeding
the reciprocal of twice the signal bandwidth.
The following results immediately follow from the Shannon sampling
theorem.
(a) Let g(x) be band-limited to kmax . Then g(x) can be recovered from its
sampled values g(n ⌬x) if
1
⌬x ⱕ (5)
2kmax
Equation (5) is known as the Nyquist sampling criterion. The largest sam-
pling interval permissible for g(x) by the Nyquist criterion is ⌬x = 1/(2kmax),
which is called the Nyquist interval.
(b) Let g(x) be support limited to 兩x兩 < W. Then its Fourier transform G(k)
can be exactly recovered from its sampled values G(n ⌬k) provided that
1
⌬k ⱕ (6)
2W
(c) A band-limited periodic function can be represented in terms of a finite
Fourier series:

冘
N

g(x) = cn e⫺i2␲nx/Wx (7)

n=⫺N

Clearly, g(x) is periodic in x with period Wx and band-limited to N/Wx (or

more precisely N/Wx ⫹ ␦ for some ␦ > 0). Since g(x) is uniquely specified
by the 2N ⫹ 1 coefficients, it is expected that 2N ⫹ 1 samples taken uni-
formly within a single period would suffice to reconstruct g(x) uniquely.
Therefore the Nyquist sampling criterion for this signal can be stated as
Wx
⌬x ⱕ (8)
2N ⫹ 1
We next consider the sampling requirements of the two data acquisition
schemes shown in Fig. 1. Note that k-space sampling is a multidimensional

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 4 Liang et al.

problem. In practice, however, one treats sampling along each dimension

separately, thus reducing it to a one-dimensional problem. Although the re-
sulting sampling pattern is not optimal, it guarantees ‘‘perfect’’ reconstruc-
tion of the underlying continuous k-space signal if infinite sampling is used.
We therefore adopt this conventional treatment to determine the require-
ments on (⌬kx , ⌬ky) and (⌬k, ⌬␾).
For rectilinear sampling, it is convenient to assume that the image
function is support limited to a central rectangle region of widths Wx and
Wy , namely,
I(x, y) = 0 for 兩x兩 ⱖ Wx /2, 兩y兩 ⱖ Wy /2 (9)
Then, according to the sampling theorem, we have

再 ⌬kx ⱕ

⌬ky ⱕ
1
Wx
1
Wy
(10)

For polar sampling, the following two standard assumptions are often
invoked [20–22]:
(a) Support-limitedness (Fig. 2a):
I(x, y) = 0 for 兹x 2 ⫹ y 2 ⱖ Rx (11)

FIGURE 2 (a) An object is bounded by a circle of radius Rx . (b) Polar

sampling of k-space.

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Introduction to Image Reconstruction 5

(b) Frequency band-limitedness (Fig. 2b):

S(kx , ky) = Ᏺ{I} = 0 for 兹k x2 ⫹ k y2 ⱖ Rk (12)
Although the first assumption is practically valid, the second
assumption is only an approximation since I(x, y) cannot be both space and
frequency limited simultaneously. These assumptions are, however, neces-
sary to enable sampling along the ␾-direction.
With assumption (a), it is easy to show that
1
⌬k ⱕ (13)
2Rx
Determining the sampling requirement along the ␾-direction is more com-
plicated. Let
Sp (k, ␾) = S(k cos ␾, k sin ␾) (14)
Clearly, Sp (k, ␾) is periodic in ␾ for a given k and can be expressed in terms
of the Fourier series as

冘
⬁

Sp (k, ␾) = cn (k)e⫺in␾ (15)

n=⫺⬁

where

冕
␲
1
cn (k) = Sp (k, ␾)e in␾ d␾ (16)
2␲ ⫺␲

The largest angular sampling interval ⌬␾ allowed for Sp (k, ␾) is determined

by the number of significant terms in the Fourier series in Eq. (15). For a
circularly symmetric object, Sp (k, ␾) is a constant over ␾ and the series will
have only a dc term. In general, the number of significant terms increases
with 兩k兩. A well-known result concerning this problem is derived in Ref. 20,
which states that Sp (k, ␾) is approximately band limited to Rx 2␲ 兩k兩 ⫹ 1 with
respect to ␾. In other words, the Fourier series coefficients cn (k) in Eq. (15)
are not significant for 兩n兩 > [Rx 2␲ 兩k兩] ⫹ 1, where the brackets represent
rounding Rx 2␲ 兩k兩 to the next higher integer. Based on the assumption of
frequency-limitedness stated in Eq. (12), cn can be ignored for any measured
k values if 兩n兩 > [2␲Rx Rk] ⫹ 1. Therefore, according to the earlier results
concerning sampling of band-limited periodic functions, the angular sam-
pling interval that satisfies the Nyquist criterion for all of the sampled k
values is given by
2␲
⌬␾ ⱕ (17)
2([2␲Rx Rk] ⫹ 1) ⫹ 1

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 6 Liang et al.

1.3 Feasible Reconstructions

The sampling theorem states that when (⌬kx , ⌬ky) or (⌬k, ⌬␾) satisfies the
Nyquist criterion, S(kx , ky) can be recovered from its sampled values and
therefore I(x, y) can be reconstructed exactly. In practice, finite coverage of
k-space is used such that S(kx , ky) is available for (kx , ky) 僆 ᏷, where ᏷ is
a finite point set. Without loss of generality, we may assume that

᏷= 再 (m ⌬kx , n ⌬ky)兩 ⫺
M
2
M N
ⱕm< ,⫺ ⱕn<
2 2
N
2
冎 (18)

for rectilinear sampling, and

᏷= 再 (m ⌬k cos n ⌬␾, m ⌬k sin n ⌬␾)兩

⫺
M
2
M N
ⱕm< ,⫺ ⱕn<
2 2
N
2
冎 (19)

for polar sampling. The image reconstruction problem can then be formally
stated as

Given S(kx , ky) for (kx , ky) 僆 ᏷

(20)
find I(x, y)

Because ᏷ is a finite point set, the reconstruction of I(x, y) is not unique

under the data-consistency constraint. In this case, it is useful to introduce
the concept of feasible reconstructions.

Definition 3 An arbitrary function Î(x, y) is said to be a feasible recon-

struction of I(x, y) if Î(x, y) is consistent with the measured data according
to the imaging equation Eq. (2). In other words, Î(x, y) is a feasible recon-
struction if

Ᏺ{I(x,
ˆ y)} = S(kx , ky) for (kx , ky) 僆 ᏷ (21)

Remark 1 The true image function I(x, y) is a feasible reconstruction.

Remark 2 If Î(x, y) is a feasible reconstruction for the reconstruction prob-

lem in Eq. (20), then Î(x, y) ⫹ e i 2␲ (p⌬kx⫹q⌬ky)⌸(x⌬kx , y⌬ky) is also a feasible
reconstruction for 兩p兩 > M/2 or 兩q兩 > N/2, where ⌸(x, y) is a two-dimensional
rectangular window function defined as

⌸(x, y) = 再 1
0
兩x兩 < 1/2
otherwise
and 兩y兩 < 1/2
(22)

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Introduction to Image Reconstruction 7

Remark 2 is readily understood noting that

冕 冕
1/2⌬kx 1/2⌬ky

e⫺i2␲ [(m⫺p)⌬kx⫹(n⫺q)⌬ky] dx dy = 0
⫺(1/2⌬kx) ⫺(1/2⌬ky)

if m ≠ p or n≠q (23)
The above two remarks indicate that for a given set of measured data, a
feasible reconstruction always exists. However, in the case of finite sam-
pling, the feasible reconstruction is not unique, and therefore additional cri-
teria need to be applied to select an image function from the many feasible
ones.

2 RECONSTRUCTION TECHNIQUES
This section describes several conventional reconstruction techniques. Ad-
vanced methods incorporating a priori constraints into the reconstruction
process are beyond the scope of the chapter. The interested reader is referred
to Refs. 23–33.

2.1 Reconstruction from Rectilinear k-Space Data

2.1.1 Reconstruction Formulas
For rectilinear sampling, we can simplify the reconstruction problem to a
one-dimensional problem based on the separability of the multidimensional
Fourier transform [4,19]. In this case, we can rewrite the imaging equation
as

冕
⬁
N N
Sn = S(n ⌬k) = I(x)e⫺i2␲n⌬kx dx ⫺ ⱕn< (24)
⫺⬁ 2 2
The following formula (commonly known as the Poisson sum formula) gov-
erns how to reconstruct I(x) from Sn :

冘 冘冉 冊
⬁ ⬁
i2␲n⌬kx 1 n
Sn e = I x⫺ (25)
n=⫺⬁ ⌬k n=⫺⬁ ⌬k
The left-hand side of Eq. (25) can be viewed as a Fourier series, with ⌬k
being the fundamental frequency and Sn being the series coefficient of the
nth harmonic. The right-hand side is a periodic extension of I(x) with period
1/⌬k.
Equation (25) can be proven as follows. First, the following equality
holds in a distribution sense:

冘 冘 冉 冊
⬁ ⬁
1 n
e i2␲n⌬kx = ␦ x⫺ (26)
n=⫺⬁ ⌬k n=⫺⬁ ⌬k

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 8 Liang et al.

Next, based on Eq. (24) we have

冘 冘 冋冕 册
⬁ ⬁ ⬁
i2␲n⌬kx
Sn e = I(x̂)e⫺i2␲n⌬kx̂ dx̂ e i2␲n⌬kx
n=⫺⬁ n=⫺⬁ ⫺⬁

冘冕
⬁ ⬁

= I(x̂)e i2␲n⌬k(x⫺x̂) dx̂

n=⫺⬁ ⫺⬁

冕 冘
⬁ ⬁

= I(x̂) e i2␲n⌬k(x⫺x̂) dx̂

⫺⬁

冕 冘 冉 冊
n=⫺⬁
⬁ ⬁
1 n
= I(x̂) ␦ x ⫺ x̂ ⫺ dx̂
⫺⬁ ⌬k ⌬k

冘冕 冉 冊
n=⫺⬁
⬁ ⬁
1 n
= I(x̂)␦ x ⫺ x̂ ⫺ dx̂
⌬k ⫺⬁ ⌬k

冘冉 冊
n=⫺⬁
⬁
1 n
= I x⫺
⌬k n=⫺⬁ ⌬k
which proves Eq. (25).
Equation (25) suggests a reconstruction formula for I(x). Assume that
I(x) is support limited to 兩x兩 < Wx /2 and that ⌬k satisfies the Nyquist criterion,
namely,
1
⌬k ⱕ (27)
Wx
Then there are no overlaps among the different components on the right-
hand side of Eq. (25). Consequently, the true image function I(x) can be
expressed in terms of Sn as

冘
⬁
1
I(x) = ⌬k Sn ei2␲n⌬kx 兩x兩 < (28)
n=⫺⬁ 2 ⌬k
Equation (28) is often referred to as the reconstruction formula for infinite
sampling. To derive a formula for the more practical case of finite sampling,
we invoke the results in the following two lemmas.
Lemma 1 Let

冘 冘
N/2⫺1

Î(x) = ⌬k Sn e i2␲n⌬kx ⫹ cn ei2␲n⌬kx (29)

n=⫺N/2 n<⫺N/2;nⱖN/2

Then Î(x) is a feasible reconstruction of I(x) with respect to the measured

data Sn , ⫺N/2 ⱕ n < N/2, for arbitrary (finite) cn .

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Introduction to Image Reconstruction 9

The lemma immediately follows from Remark 2, which indicates that the
data-consistency constraints allow the unmeasured data to take finite arbi-
trary values in the reconstruction step. Therefore an important practical ques-
tion is what values we should assign to the cn . Clearly the answer is appli-
cation dependent. In the conventional reconstruction technique, the cn is
selected based on the minimum-norm constraint. As a result, the unmeasured
data are forced to be zero as indicated by the following lemma.
Lemma 2 Let Î(x) be a feasible reconstruction given in Eq. (29). Then

冕
1/2⌬k

c̄n = arg min

c
兩I(x)兩
ˆ 2 dx = 0 (30)
n
⫺(1/2⌬k)

The lemma can be easily justified by invoking the Parseval theorem [19],

冕 冘 冘
1/2⌬k N/2⫺1
1
兩I(x)兩
ˆ 2 dx = 兩⌬kSn兩2 ⫹ 兩cn兩2 (31)
⌬k ⫺(1/2⌬k) n=⫺N/2 n<⫺N/2;nⱖN/2

which reaches the minimum when cn = 0.

Based on the above two lemmas, the minimum-norm feasible recon-
struction can be expressed as

冘
N/2⫺1
1
Î(x) = ⌬k Sn e i2␲n⌬kx 兩x兩 < (32)
n=⫺N/2 2⌬k
Equation (32) is popularly known as the Fourier reconstruction formula,
which can be evaluated using a fast Fourier transform algorithm [34].
Remark 3 Extending Eq. (32) to two dimensions yields

冘 冘
M/2⫺1 N/2⫺1

Î(x, y) = ⌬kx ⌬ky S(m ⌬kx , n ⌬ky)e i2␲ (m⌬kx x⫹n⌬ky y) (33)
m=⫺M/2 n=⫺N/2

for 兩x兩 < 1/2 ⌬kx and 兩y兩 < 1/2 ⌬ky .
2.1.2 Image Characteristics
Clearly, the Fourier reconstruction given by Eq. (32) is not identical to the
true image function I(x). The following lemma relates Î(x) to I(x).
Lemma 3 The Fourier reconstruction Î(x) given by Eq. (32) is related to
the true image function I(x) by the convolution equation

冕
⬁

Î(x) = I(x̂)h(x ⫺ x̂) dx̂ (34)

⫺⬁

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 10 Liang et al.

where h(x), known as the point spread function (PSF), is given by

sin(␲ N ⌬kx) ⫺i␲⌬kx
h(x) = ⌬k e (35)
sin(␲ ⌬kx)
This above lemma can be justified by directly deriving the PSF asso-
ciated with the Fourier reconstruction formula. Specifically, setting I(x) =
␦ (x), the measured data are Sn = 1. Substituting these values into the Fourier
reconstruction formula gives

冘
N/2⫺1

h(x) = ⌬k e i2␲n⌬kx (36)

n=⫺N/2

Further simplification yields Eq. (35) immediately.

Remark 4 The phase term e⫺i␲ ⌬kx in h(x) can be eliminated, if necessary,
by covering k-space from ⫺(N/2) ⌬k to (N/2) ⌬k instead of from ⫺(N/2) ⌬k
to ((N ⫺ 1)/2) ⌬k. However, this term is often ignored, because the phase
variation across a pixel is (␲/N), which is insignificant in practice.
Note that h(x) is a periodic function, and within each period it displays
characteristics similar to those of a sinc function. The width of its main
lobe, as measured by the interval between the first two zero crossings, is
2/(N ⌬k). The effective width Wh of h(x) is often taken to be the width of
an approximating rectangular pulse with height h(0) and the same area. It
is easy to show that

冕
1/2⌬k
1 1
Wh = h(x) dx = (37)
h(0) ⫺(1/2⌬k) N ⌬k
which is exactly half the width of the main lobe of h(x).
The right-hand side of Eq. (37) is known as the Fourier pixel size and
denoted as ⌬xF in contrast to the usual image pixel size ⌬x. Note that ⌬x
can be made arbitrarily small using any signal interpolation scheme, but
image resolution is fundamentally limited to ⌬xF . Therefore, in order for
two point sources to be distinguishable in Î(x), their separation has to be
larger than ⌬xF . Another implication of Eq. (37) is that Wh and N cannot be
reduced simultaneously; in other words, improving image resolution and
reducing the number of measured data points cannot be achieved simulta-
neously.
In addition to the loss of resolution in Î(x), the convolution operation
in Eq. (34) also results in the well-known Gibbs ringing artifact in Î(x) when
I(x) contains step discontinuities, as illustrated in Fig. 3. A full theoretical

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Introduction to Image Reconstruction 11

FIGURE 3 Illustration of the Gibbs phenomenon.

account of the Gibbs phenomenon is not very difficult but neither is it very
interesting. We shall therefore limit ourselves to describing its source and
behavior.
The Gibbs phenomenon is fundamentally related to the convergence
behavior of the Fourier series [19]. It is well known that when I(x) is a
smooth function, Î(x), given in Eq. (32), uniformly converges to I(x) as
N → ⬁. This is not the case when I(x) contains step discontinuities. Assume
that I(x) has a step discontinuity at x = x0 such that I(x ⫹ ⫺
0 ) > I(x 0 ). It can be
shown that the maximum overshoot location x̃ N in Î(x) approaches x 0⫹ as
⫹

N → ⬁, and that
ˆ ˜⫹
lim [I(x ⫹ ⫹ ⫺
N ) ⫺ I(x0 )] ⬇ 9%[I(x 0 ) ⫺ I(x 0 )] (38)
N→ ⬁

Similarly, for the maximum undershoot, we have

ˆ ˜⫺
lim [I(x ⫺ ⫹ ⫺
N ) ⫺ I(x0 )] ⬇ ⫺9%[I(x 0 ) ⫺ I(x 0 )] (39)
N→ ⬁

Therefore, as N → ⬁, the value of the maximum overshoot (undershoot)

does not tend to zero but instead tends to a finite value. The existence of
this finite, nonzero, limiting value of the overshoot (undershoot) is due to
the nonuniform convergence of Î(x) to I(x) in the vicinity of the discontin-
uous points of I(x). This nonuniform convergence behavior of the limit Î(x)
→ I(x) as N → ⬁ is formally called the Gibbs phenomenon.

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 12 Liang et al.

The Gibbs ringing artifact is a common image distortion that exists in

Fourier images, which manifests itself as spurious ringing around sharp
edges, as illustrated in Fig. 3. The maximum undershoot or overshoot of the
spurious ringing is about 9% of the intensity discontinuity and is indepen-
dent of the number of data points used in the reconstruction. The frequency
of oscillation, however, increases as more data points are used. For this
reason, when a large number of data points is used in practice, the spurious
ringing does not cover an appreciable distance in the reconstructed image
and thus becomes invisible.
A straightforward method to reduce the Gibbs ringing artifact is to
filter the measured data before the reconstruction algorithm is applied. This
operation, known as windowed Fourier reconstruction, is described by

冘
N/2⫺1

Î(x) = ⌬k wn Sn e i2␲n⌬kx (40)

n=⫺N/2

where wn is a preselected filter (or window) function. The PSF associated

with this reconstruction formula is

冘
N/2⫺1

h(x) = ⌬k wn e i2␲n⌬kx (41)

n=⫺N/2

By choosing wn appropriately, one can significantly suppress the oscillations

in h(x) and thus the Gibbs ringing in Î(x). A variety of window functions
have been proposed for this purpose (see Ref. 35 for a comprehensive re-
view). The popular Hamming window is defined as

wn = 0.54 ⫹ 0.46 cos(2␲ n/N) ⫺N/2 ⱕ n < N/2 (42)

and its effects on Gibbs ringing reduction are illustrated in Fig. 4.

Although the windowing approach is effective in suppressing the
Gibbs ringing, it is at the price of spatial resolution, as illustrated in Fig. 4.
This property can be understood by examining the effective width of the
associated PSF. Specifically,

冕 冘
1/2⌬k N/2⫺1
1

冘
Wh = N/2⫺1 ⌬k wm e i2␲m⌬kx dx
⫺(1/2⌬k) m=⫺N/2
⌬k wm
m=⫺N/2

冘
= N/2⫺1

(wm /w0) ⌬k
m=⫺N/2

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Introduction to Image Reconstruction 13

FIGURE 4 Phantom images reconstructed using the Fourier method with

256 data points along the vertical direction. The horizontal direction was
reconstructed using (a) 256 data points, (b) 64 data points, and (c) 64
data points weighted by the Hamming window function.

Since w0 ⱖ wm for any practical window function used for this purpose, we
have

1
Wh ⱖ (43)
N ⌬k

More effective filtering methods for Gibbs ringing reduction are available;
the interested reader is referred to Refs. 33 and 36–38.

2.2 Reconstruction from Polar k-Space Data

A conventional way to reconstruct an image from polar k-space data is to
use the well-known filtered back-projection algorithm.2 Before describing
this algorithm, it is useful to introduce formally the Radon transform, al-
though MRI does not collect data directly in Radon space per se.
2.2.1 Two-Dimensional Radon Transforms
The two-dimensional Radon transform of I(x, y) is defined as [42–44]

P(s, ␾) = ᏾{I(x, y)} = 冕 L

I(x, y) dl (44)

2
Interpolation-based Fourier reconstruction methods [39–41] are discussed in later chapters
of the book.

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 14 Liang et al.

where the integral path L is defined by

x cos ␾ ⫹ y sin ␾ = s (45)

Alternatively, it can be written as

冕冕
⬁ ⬁

P(s, ␾) = I(x, y)␦ (x cos ␾ ⫹ y sin ␾ ⫺ s) dx dy (46)

⫺⬁ ⫺⬁

For a fixed ␾, P(s, ␾) as a function of s is a projection of I(x, y) along L

(Fig. 5), and ␾ is referred to as the projection angle.
The Radon transform is closely related to the Fourier transform by the
well-known projection-slice theorem, as described below.

Theorem 2 (Projection-Slice Theorem) Let P(s, ␾) be the Radon trans-

form of I(x, y) and S(kx , ky) be its Fourier transform. For a fixed ␾0 , the one-
dimensional Fourier transform of P(s, ␾0) along the s-axis is identical to
S(kx , ky) evaluated along a straight line specified by ky = kx tan ␾0 . The
theorem can be expressed in compact form as

Ᏺs{P(s, ␾0)} = S(k cos ␾0 , k sin ␾0) (47)

where Ᏺs denotes one-dimensional Fourier transform along the s-axis.

FIGURE 5 Two-dimensional Radon transform as line integrals.

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Introduction to Image Reconstruction 15

The projection-slice theorem can be proven as follows:

冕
⬁

Ᏺs{P(s, ␾0)} = P(s, ␾0)e⫺i2␲ks ds

⫺⬁

冕 冋冕 冕 册
⬁ ⬁ ⬁

= I(x, y)␦ (x cos ␾0 ⫹ y sin ␾0 ⫺ s) dx dy e⫺i2␲ks ds

⫺⬁ ⫺⬁ ⫺⬁

冕 冕 冋冕 册
⬁ ⬁ ⬁

= I(x, y) ␦ (x cos ␾0 ⫹ y sin ␾0 ⫺ s)e⫺i2␲ks ds dx dy

⫺⬁ ⫺⬁ ⫺⬁

冕冕
⬁ ⬁

= I(x, y)e⫺i2␲k(x cos ␾0⫹y sin ␾0) dx dy

⫺⬁ ⫺⬁

冕冕
⬁ ⬁

= I(x, y)e⫺i2␲(xk cos ␾0⫹yk sin ␾0) dx dy

⫺⬁ ⫺⬁

= S(k cos ␾0 , k sin ␾0)

The inverse Radon transform is given in the following theorem.

Theorem 3 (Inverse Radon Transform) Let P(s, ␾) be the Radon trans-
form of I(x, y) defined in Eq. (46). Then I(x, y) can be reconstructed from
P(s, ␾) by
⭸P(s, ␾)

冕冕
␲ ⬁
1 ⭸s
I(x, y) = ds d␾ (48)
2␲ 2 0 ⫺⬁ x cos ␾ ⫹ y sin ␾ ⫺ s
Proof We first invoke the two-dimensional inverse Fourier transform

冕冕
⬁ ⬁

I(x, y) = S(kx , ky)ei2␲ (kxx⫹kyy) dkx dky (49)

⫺⬁ ⫺⬁

Converting it to the polar coordinate system (using kx = k cos ␾, ky =

k sin ␾) yields

冕冕
2␲ ⬁

I(x, y) = Sp (k, ␾)ei2␲k(x cos ␾⫹y sin ␾)k dk d␾ (50)

0 0

where Sp (k, ␾) = S(k cos ␾, k sin ␾). Making use of the symmetry that
Sp (k, ␾) = Sp (⫺k, ␾ ⫾ ␲), Eq. (50) can be rewritten as

冕冕
␲ ⬁

I(x, y) = 兩k兩Sp (k, ␾)ei2␲k(x cos ␾⫹y sin ␾) dk d␾ (51)

0 ⫺⬁

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 16 Liang et al.

The inner integral is simply the inverse Fourier transform of 兩k兩Sp (k, ␾)
evaluated at s = x cos ␾ ⫹ y sin ␾. Based on the convolution and derivative
properties of the Fourier transform and the projection-slice theorem, we have

Ᏺ⫺1{兩k兩Sp (k, ␾)} = Ᏺ⫺1{kSp (k, ␾)sgn(k)}

= Ᏺ⫺1{kSp (k, ␾)} * Ᏺ⫺1{sgn(k)}
1 ⭸ 1
= P(s, ␾) *
i2␲ ⭸s ⫺i␲s
1 ⭸P(s, ␾) 1
= * s (52)
2␲ 2 ⭸s

Substituting Eq. (52) into Eq. (51) proves Eq. (48).

2.2.2 Filtered Back-Projection Reconstruction

The inversion formula in Eq. (48) may be interpreted as consisting of two
steps. The first step is the operation in the inner integral, which filters each
projection P(s, ␾) with a one-dimensional filter whose frequency response
is 兩k兩. More specifically, according to Eq. (52), the filtered projection denoted
as P̂(s, ␾) can be expressed as

P̂(s, ␾) = Ᏺ⫺1{兩k兩Sp (k, ␾)} (53)

or equivalently

1 ⭸P(s, ␾) 1
P̂(s, ␾) = * s (54)
2␲ 2 ⭸s

The second step in the inversion formula is to back-project each filtered

projection P̂(s, ␾). The back-projection operator Ꮾ is defined in two di-
mensions as

冕
␲

Ꮾ{P(s,
ˆ ␾)} = ˆ cos ␾ ⫹ y sin ␾, ␾) d␾
P(x (55)
0

The term back-projection can be understood from the fact that the operation
P̂(s, ␾) → P̂(x cos ␾ ⫹ y sin ␾, ␾) in essence assigns the value of each
point in P̂(s, ␾) to a line defined by s = x cos ␾ ⫹ y sin ␾.
The popular filtered back-projection reconstruction algorithm is an im-
plementation of the above two steps. First, for a given set of polar k-space
data Sp (k, ␾), filtered projections are generated according to Eq. (53). To
avoid noise amplification by the 兩k兩 filter in this equation, an approximate

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Introduction to Image Reconstruction 17

filter function H(k) is often used in practice. Consequently, the filtered pro-
jections are obtained by
P̂(s, ␾) = Ᏺ⫺1{H(k)Sp (k, ␾)} (56)
Some commonly used filter functions are
All-pass filter:
H(k) = 1 (57)
Ram-Lak filter:

H(k) = 兩k兩 ⌸ 冉冊
k
Wk
(58)

Shepp-Logan filter:

H(k) = 兩k兩sinc 冉冊 冉冊
␲k
Wk
⌸
k
Wk
(59)

Low-pass cosine filter:

H(k) = 兩k兩cos 冉冊 冉冊
␲k
Wk
⌸
k
Wk
(60)

Generalized Hamming filter:

H(k) = 兩k兩 冋 0.54 ⫹ 0.46 cos 冉 冊册 冉 冊

2␲k
Wk
⌸
k
Wk
(61)

where ⌸(⭈) is the unit rectangular window function and Wk denotes the
desired frequency bandwidth of the filtered projections. After a filter function
is selected, P̂(s, ␾) can be calculated using the Fourier reconstruction al-
gorithm described previously.
We next consider the practical implementation of the back-projection
step. For projections taken at discrete angles ␾n = n ⌬␾, a back-projection
image Î(x, y) from P̂(s, ␾n) is calculated using the following approximation
to Eq. (55):

冘
N␾⫺1

ˆ y) = ⌬␾
I(x, ˆ cos ␾n ⫹ y sin ␾n , ␾n)
P(x (62)
n=0

where N␾ denotes the total number of projections measured. In practice,

P̂(s, ␾) are sampled both angularly and radially so that interpolation on
P̂(s, ␾) along the radial direction is necessary. Assume that P̂(s, ␾) is avail-
able for s = n ⌬s with ⫺(Ns /2) ⱕ n < (Ns /2) and ␾ = n ⌬␾ with n = 0, 1,
. . . , N␾ ⫺ 1, and that Î(x, y) is evaluated on a rectangular grid (x = nx ⌬x,

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 18 Liang et al.

y = ny ⌬y). Then, in evaluating Eq. (62), we will encounter the case that
P̂(nx ⌬x cos ␾n ⫹ ny ⌬y sin ␾n , ␾n) is not defined because
Ns Ns
s = nx ⌬x cos ␾n ⫹ ny ⌬y sin ␾n ≠ m ⌬s ⫺ ⱕmⱕ ⫺1 (63)
2 2
In this case, P̂(s, ␾n) is estimated using an appropriate signal interpolation
scheme. For example, with nearest neighbor or zero-order interpolation, we
have
ˆ
P(s, ␾n) = P(m
ˆ 0 ⌬s, ␾n) (64)
where
m0 = arg min兩s ⫺ m ⌬s兩 (65)
m

Higher-order interpolation methods are often more accurate but also com-
putationally more expensive. For example, in the linear interpolation
method, we select m0 such that m0 ⌬s ⱕ s < (m0 ⫹ 1) ⌬s and estimate P̂(s,
␾n) by
(m0 ⫹ 1) ⌬s ⫺ s ˆ
ˆ
P(s, ␾n) = P(m0 ⌬s, ␾n)
⌬s
s ⫺ m0 ⌬s ˆ
⫹ P[(m0 ⫹ 1) ⌬s, ␾n ] (66)
⌬s
The quality of filtered back-projection reconstructions is dependent on
a number of factors, most notably the total number of projections (or radial
k-space lines) collected and the filter function H(k) used. Some examples of
reconstruction demonstrating the effect of these factors are shown in Fig. 6.
A special result regarding the filter function is summarized in the following
remark.
Remark 5 When H(k) is an all-pass filter, P̂(s, ␾) = P(s, ␾), and the
resulting reconstruction algorithm is known as the back-projection algorithm.
Let Î(x, y) be the back-projection image, namely,

冕
␲

Î(x, y) = Ꮾ{P(s, ␾)} = P(x cos ␾ ⫹ y sin ␾, ␾) d␾ (67)

0

Then Î(x, y) is related to I(x, y) by

1
Î(x, y) = I(x, y) ** 2 (68)
x ⫹ y2
where ** denotes two-dimensional convolution.

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Introduction to Image Reconstruction 19

FIGURE 6 Filtered back-projections of a phantom image. Images (a) and

(b) were reconstructed using an all-pass filter, while images (c) and (d)
were reconstructed using the ‘‘Ram-Lak’’ filter. Images (a) and (c) were
obtained from 256 projections each with 256 data points, and images (b)
and (d) were obtained from 64 projections each with 256 data points.

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 20 Liang et al.

Equation (68) can be derived as follows:

冕
␲

Î(x, y) = P(x cos ␾ ⫹ y sin ␾, ␾) d␾

0

冕冕
␲ ⬁

= Sp (k, ␾)e i2␲k(x cos ␾⫹y sin ␾) dk d␾

0 ⫺⬁

冕冕
2␲ ⬁

= Sp (k, ␾)e i2␲k(x cos ␾⫹y sin ␾) dk d␾

0 0

冕冕
2␲ ⬁

= k⫺1Sp (k, ␾)e i2␲k(x cos ␾⫹y sin ␾)k dk d␾

冕冕
0 0
⬁ ⬁
1
= Sp (kx , ky)e i2␲(xkx⫹yky) dkx dky
兹k ⫹ k y
22
⫺⬁ ⫺⬁

再 冎
x

1
= Ᏺ⫺1 ** I(x, y)
兹k ⫹ k y
2 2
x

1
= 2 **
I(x, y)
兹x ⫹ y
2

where use of the following identity is made:

Ᏺ⫺1 再 兹k ⫹ k2
x
1
2
y
冎 =
1
兹x ⫹ y 2
2
(69)

3 CONCLUSION
Image reconstruction is an essential step in MR imaging because spatial
information is encoded into the measured data. Depending on how spatial
information is encoded into the measured data, the image reconstruction
technique can vary considerably. This chapter provides an introductory dis-
cussion of two fundamental problems: (a) reconstruction from rectilinear k-
space data and (b) reconstruction from polar k-space data. Many practical
MRI data acquisition schemes lend themselves naturally to one of these two
reconstruction problems. For imaging schemes using other k-space coverage,
signal interpolation is often used to regrid the measured data into one of
these two standard formats so that the basic reconstruction algorithms can
be applied.
This chapter focuses on some of the fundamental mathematical con-
cepts in image reconstruction. Many important issues, such as resolution
limitation, noise characteristics, signal interpolation, use of a priori con-

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Introduction to Image Reconstruction 21

straints, and computational efficiency are not dealt with in any depth. Inter-
ested readers are referred to Refs. 4, 23–33, and 45–52.

ACKNOWLEDGMENTS
This work was partially supported by the following grants: NSF-BES-95-
02121, NIH-RO1-CA51430, NIH-P41-RR05964, and NIH-R21-HL062336.

REFERENCES
1. PC Lauterbur. Image formation by induced local interactions: examples em-
ploying nuclear magnetic resonance. Nature 242:190–191, 1973.
2. WS Hinshaw, AH Lent. An introduction to NMR imaging: from the Bloch
equation to the imaging equation. Proc IEEE 71:338–350, 1983.
3. EM Haacke, RW Brown, MR Thompson, R Venkatesan. Magnetic Resonance
Imaging: Physical Principles and Sequence Design. John Wiley, New York,
1999.
4. Z-P Liang, PC Lauterbur. Principles of Magnetic Resonance Imaging: A Signal
Processing Perspective. IEEE Press, Piscataway, NJ, 1999.
5. ZH Cho, HS Kim, HB Song, J Cumming. Fourier transform nuclear magnetic
resonance tomographic imaging. Proc IEEE 70:1152–1173, 1982.
6. GH Glover, JM Pauly. Projection reconstruction techniques for reduction of
motion artifacts. Magn Reson Med 28:275–289, 1992.
7. L Bolinger, J Leigh. Hadamard spectroscopic imaging (HSI) for multivolume
localization. J Magn Reson 80:162–167, 1988.
8. DM Healy, JB Weaver. Two applications of wavelet transforms in magnetic
resonance imaging. IEEE Trans Info Theory 38:840–860, 1992.
9. LP Panych, PD Jakab, FA Jolesz. An implementation of wavelet encoded MRI.
J Magn Reson Imag 3:649–655, 1993.
10. LP Panych, FA Jolesz. A dynamically adaptive imaging algorithm for wavelet
encoded MRI. Magn Reson Med 32:738–748, 1994.
11. LP Panych. Theoretical comparison of Fourier wavelet encoding in magnetic
resonance imaging. IEEE Trans Med Imag 15:141–153, 1996.
12. RD Peters, ML Wood. Multilevel wavelet-transform encoding in MRI. J Magn
Reson Imag 6:529–540, 1996.
13. Y Cao, DN Levin. Feature-recognizing MRI. Magn Reson Med 30:305–317,
1993.
14. Y Cao, DN Levin. On the relationship between feature-recognizing MRI and
MRI encoded by singular value decomposition. Magn Reson Med 33:140–142,
1995.
15. GP Zientara, LP Panych, FA Jolesz. Dynamically adaptive MRI with encoding
by singular value decomposition. Magn Reson Med 32:268–274, 1994.
16. LP Panych, C Oesterle, GP Zientara, J Hennig. Implementation of fast gradient-
echo SVD encoding technique for dynamic imaging. Magn Reson Med 35:
554–562, 1995.

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 22 Liang et al.

17. RN Bracewell. The Fourier Transform and Its Applications. McGraw-Hill, New
York, 1986.
18. RJ Marks II. Introduction to Shannon Sampling and Interpolation Theory.
Springer-Verlag, New York, 1991.
19. A Papoulis. Signal Analysis. McGraw-Hill, New York, 1977.
20. PA Rattey, AG Lindgren. Sampling the 2-D Radon transform. IEEE Trans
Acoust Speech Signal Process ASSP-29:994–1002, 1994.
21. H Stark. Sampling theorems in polar coordinates. J Opt Soc Am 69:1519–
1525, 1979.
22. H Stark, M Wengrovitz. Comments and corrections on the use of polar sam-
pling theorems in CT. IEEE Trans Acoust Speech Signal Process ASSP-31:
1329–1331, 1983.
23. Z-P Liang, F Boada, T Constable, EM Haacke, PC Lauterbur, MR Smith.
Constrained reconstruction methods in MR imaging. Rev Magn Reson Med 4:
67–185, 1992.
24. Z-P Liang, PC Lauterbur. Constrained imaging: overcoming the limitations of
the Fourier series. IEEE EMB Magazine 15:126–132, 1996.
25. Z-P Liang, PC Lauterbur. A generalized series approach to MR spectroscopic
imaging. IEEE Trans Med Imag MI-10:132–137, 1991.
26. Z-P Liang, PC Lauterbur. Efficient magnetic resonance imaging using a gen-
eralized series model. IEEE Trans Med Imag 13:677–686, 1994.
27. Z-P Liang, H Jiang, CP Hess, PC Lauterbur. Dynamic imaging by model es-
timation. Int J Imag Syst Techn 8:551–557, 1997.
28. EM Haacke. Solving for nonideal conditions in two-dimensional Fourier trans-
form magnetic resonance imaging using a generalized inverse transform. In-
verse Problems 3:421–435, 1987.
29. EM Haacke, Z-P Liang, SH Izen. Constrained reconstruction: a superresolution
optimal signal-to-noise alternative to the Fourier transform. Med Phys 16:388–
397, 1989.
30. JM Hanson, Z-P Liang, EC Wiener, PC Lauterbur. Fast dynamic imaging using
two reference images. Magn Reson Med 36:172–175, 1996.
31. SK Plevritis, A Macovski. Spectral extrapolation of spatially bounded images.
IEEE Trans Med Imag MI-14:487–497, 1995.
32. S Gull, J Daniel. Image reconstruction from incomplete and noisy data. Nature
272:686–690, 1978.
33. RT Constable, RM Henkelman. Data extrapolation for truncation artifact re-
moval. Magn Reson Med 17:108–118, 1991.
34. EO Brigham. The Fast Fourier Transform. Prentice-Hall, Englewood Cliffs,
NJ, 1974.
35. FJ Harris. On the use of windows for harmonic analysis with the discrete
Fourier transform. Proc IEEE 66:51–83, 1978.
36. S Amartur, EM Haacke. Modified iterative model based on data extrapolation
method to reduce Gibbs ringing. J Magn Res Imag 1:307–317, 1991.
37. S Amartur, Z-P Liang, F Boada, EM Haacke. A phase constrained data extrap-
olation method to reduce Gibbs ringing. J Magn Reson Imag 1:721–724, 1991.

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Introduction to Image Reconstruction 23

38. DL Parker, GT Gullberg, PR Frederick. Gibbs artifact removal in magnetic

resonance imaging. Med Phys 14:640–645, 1987.
39. H Choi, DC Munson Jr. Direct-Fourier reconstruction in tomography and syn-
thetic aperture radar. Int J Imag Syst Techn 9:1–13, 1998.
40. H Stark, JW Woods, I Paul, R Hingorani. Direct Fourier reconstruction in
computer tomography. IEEE Trans Acoust Speech Signal Process ASSP-29:
237–245, 1981.
41. H Stark, JW Woods, I Paul, R Hingorani. An investigation of computerized
tomography by direct Fourier inversion and optimum interpolation. IEEE Trans
Biomed Eng BME-29:496–505, 1981.
42. SR Deans. The Radon Transform and Some of Its Applications. John Wiley,
New York, 1983.
43. GT Herman. Image Reconstruction from Projection: The Fundamentals of
Computerized Tomography. Academic Press, New York, 1980.
44. AC Kak, M Slaney. Principles of Computerized Tomographic Imaging. IEEE
Press, Piscataway, NJ, 1988.
45. M Bertero, P Boccacci. Introduction to Inverse Problems in Imaging. Institute
of Physics Publishing, Bristol and Philadelphia, 1998.
46. Y Cao, DN Levin. Imaging with spatially variable resolutions. J Magn Reson
Imag 2:701–709, 1992.
47. Y Cao, DN Levin, L Yao. Locally focused MRI. Magn Reson Med 34:858–
867, 1995.
48. Y Cao, DN Levin. Using an image database to constrain the acquisition and
reconstruction of MR images of the human brain. IEEE Trans Med Imag 14:
350–361, 1995.
49. Y Cao, DN Levin. Using prior knowledge of human anatomy to constrain MR
image acquisition and reconstruction. Magn Reson Imag 15:669–677, 1997.
50. DL Parker, GT Gullberg. Signal-to-noise efficiency in magnetic resonance im-
aging. Med Phys 17:250–257, 1990.
51. X Hu, T Parrish. Reduction of field of view for dynamic imaging. Magn Reson
Med 31:691–694, 1994.
52. H Sedarat, AB Kerr, JM Pauly, DG Nishimura. Partial-FOV reconstruction in
dynamic spiral imaging. Magn Reson Med 43:429–439, 2000.

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 2
Wavelet-Based Multiresolution
Local Tomography

F. Rashid-Farrokhi
Lucent Technologies, Holmdel, New Jersey

K. J. Ray Liu
University of Maryland, College Park, Maryland

1 INTRODUCTION
The reconstruction of a wavelet transform of an image directly from the
Radon transform of the image has been of great interest recently [1–15].
Some applications of this technique include but are not limited to image
enhancement, feature extraction directly from the Radon transform domain,
and local tomography. The purpose of this chapter is to present methods of
extracting wavelet transform coefficients directly from the Radon transform
data and to show that wavelet transform methods are useful tools in the
study of local and semilocal tomography. We also show that the various
image analysis methods that employ wavelet transforms or other multiscale
analyses intertwine with Radon transform inversion. This allows for the pos-
sibility of performing image enhancement or feature extraction directly on
tomographic data.

25

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 26 Rashid-Farrokhi and Liu

It is well known that in dimension two, and in fact in any even di-
mension, the Radon transform is not local, that is, the recovery of an image
at any fixed point requires the knowledge of all projections of the image.
This means that a patient would have to be exposed to a relatively large
amount of x-rays even if it was desired to view only a small part of the
patient’s body. Interesting techniques have been presented in Refs. 7–13 to
reduce exposure, and at the same time to enable the perfect reconstruction
of a region of interest.
The application of wavelet theory to the inversion of the Radon trans-
forms was first proposed and formulated by Walnut [1] and later was im-
plemented by DeStefano and Olson [9]. This formula was based on an in-
tertwining between the one-dimensional continuous wavelet transform of the
projection data at each angle and the two-dimensional wavelet transform of
the original image. The fundamental observation was that the admissibility
or vanishing moment condition that is characteristic of a wavelet is preserved
under the Hilbert transform. Moreover [2,9], the Hilbert transform of a func-
tion with many vanishing moments should decay very rapidly. This is related
to the fact that certain singular integral operators are almost diagonalized
by wavelets [23]. Bernstein and Walnut [2] used the intertwining formula,
developed by Walnut [1], for local recovery and obtained explicit error es-
timates on the recovered image within the region of interest. It was noted
that high frequency features of an image can be recovered locally using the
wavelet transform.
The first numerical algorithm using wavelets for local reconstruction
was implemented by DeStefano and Olson [9]. This algorithm reconstructs
the local values of a function f directly from the one-dimensional wavelet
transform of R␪ f at each angle ␪. Delaney and Bresler [10] compute the
two-dimensional separable wavelet transform of a function directly from the
projection data as a means of carrying out local recovery from local mea-
surements. Both algorithms take advantage of the observation that the Hil-
bert transform of a function with many vanishing moments has rapid decay;
and both algorithms recover the high-resolution parts of the image locally
(that is, by exposing the region of interest plus a small extra margin) and
obtain the low-resolution parts by global measurements at a few angles. Both
of these algorithms exhibit similar savings in exposure and a similar quality
of the reconstructed image in the region of interest. Olson [13] has improved
the algorithm developed by DeStefano and Olson [9] by replacing the usual
wavelet transform with the local trigonometric transform of Coifman and
Meyer [17] and has reduced the exposure still further. The algorithms of
DeStefano and Olson [9], Delaney and Bresler [10], and Olson [13] are

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Wavelet-Based Multiresolution Local Tomography 27

called pseudo local tomography algorithms in that they use measurements

far from the region of interest at a sparse set of angles to recover the function
exactly in the region of interest.
Rashid-Farrokhi et al. [14] devised a local wavelet–based algorithm
to reconstruct an approximation of the low-resolution parts of the image as
well as the high-resolution parts using only local measurements. The algo-
rithm is based on the observation that in some cases, the Hilbert transform
of a compactly supported scaling function also has essentially the same
support as the scaling function itself. These phenomena are related to the
number of vanishing moments of the scaling function of an orthonormal or
biorthonormal wavelet basis. That is, if ␾ (t) is such a scaling function, and
if ␾ˆ ( j ) (0) = 0 for j = 1, 2, . . . , K, for some large K, then the Hilbert
transform of ␾ will have rapid decay. An extension of the algorithms in
Refs. 10–14 to fan-beam tomography was presented by Rashid-Farrokhi et
al. [15].
The goal of the algorithms in Ref. 14 is to reconstruct the function
locally from local measurements up to the null space of the interior Radon
transform. That is, the problem of recovery of local values of a function
from local projections only is not uniquely solvable [25–28]. Natterer [26]
presented an example of functions that are nonzero on a disk but whose
projections on all lines intersecting that disk are zero. Such a function is
said to be an element of the null space of the interior Radon transform. Any
algorithm that uses only local measurements cannot reconstruct these null
space elements. The advantage is that taking only local measurements is
much easier to implement in hardware. It has been shown that the elements
of the null space of the interior Radon transform do not vary much in the
region of interest [26].
Another local tomography technique is ⌳-tomography, which is used
to reconstruct the function ⌳f ⫺ ␮⌳⫺1f rather than the density function f
[6,7]. The function ⌳f has the same singularities as f and is cupped where
f is constant. The addition of the cup correction factor ␮⌳⫺1f results in good
qualitative reconstructions of f [8].
This chapter is organized as follows: In Section 2 we will briefly in-
troduce the Radon transform and discuss the nonlocality of the Radon trans-
form and the conventional reconstruction technique, i.e., the filtered back-
projection method. In Section 3, after reviewing the basics of the wavelet
transform, we will introduce a full-data reconstruction technique based on
the wavelet transform. We will discuss the locality property of the proposed
algorithm in Section 4. Section 5 then discusses the implementation of these
methods. The extension of the above methods to fan-beam geometry is pre-
sented in Section 6.

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 28 Rashid-Farrokhi and Liu

2 PRELIMINARIES AND NOTATIONS

In this section, we will briefly introduce the terminology and definitions
required in the subsequent discussions. In this chapter we use the following
notations: the d-dimensional Euclidean space is denoted by R d. Given a set
S 傺 R d, 1S denotes the indicator function of S. We define the Fourier trans-
→ →
) = 兰R d f( x )e j2␲␻ ⭈ x d x. The inverse Fourier transform is
→ → →
form in R d by ˆf(␻ → →
→ ∨ → → ⫺j 2␲␻ ⭈ x →
defined by f( x ) = ( f ) ( x ) = 兰R d f(␻)e
ˆ ˆ dw. Both continuous and dis-
crete convolution operators are denoted by *.

2.1 Radon Transform

In computerized tomography (CT), a cross section of the human body is
scanned by a nondiffracting thin x-ray beam whose intensity loss is recorded
by a set of detectors. The Radon transform (RT) is a mathematical tool that
is used to describe the recorded intensity losses as averages of the tissue
density function over hyperplanes, which in dimension two are lines. Given
f( →x ), restricted to a disc of radius one, we define the Radon transform of
f by

R␪ f(s) = 冕 → →
x ⭈ ␪ =s
f( →x ) d →x = 冕
␪⬜
→
f(s␪ ⫹ y) dy

→
where ␪ = (cos→
␪, sin ␪), ␪ 僆 [0, 2␲), s 僆 R, and ␪ ⬜ is the subspace per-
pendicular to ␪.
The interior Radon transform [27,28] is the Radon transform restricted
to lines passing through the region of interest (ROI), which is a circle of
radius r (r < 1) about the origin. It is defined by
R␪ f(s) → R␪ f(s)1[⫺r,r] (s)
The problem of recovery of f from the interior Radon transform is called
the interior problem or region of interest tomography. The interior problem
in dimension two is not uniquely solvable, i.e., there are functions that are
not zero in the region of interest but whose projections on lines intersecting
that region are zero. However, these functions do not vary much inside the
region of interest, and in fact a crude approximation to the missing projec-
tions suffices to approximate f well inside the region of interest up to an
additive constant [26].

2.2 Reconstruction
The basic formula for inverting the Radon transform is based on the fact
that the Fourier transform of the Radon transform with respect to the variable

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

 Wavelet-Based Multiresolution Local Tomography 29

s is the Fourier transform of the function f along a line passing through the
origin. This property is known as the projection theorem or Fourier slice
theorem:
ˆ
(R ˆ →
␪ f )(␻) = f(␻␪ ) ␻僆R
Thus the Fourier transform of the projections at enough angles could in
principle be assembled into a complete description of the two-dimensional
Fourier transform of the image and then simply inverted to arrive at the
function f. Using the polar Fourier inversion formula and the Fourier slice
theorem, we can reconstruct the function f from the projection data R␪ f(s)
by

冕冕
␲ ⬁
→ →
→
f( x ) = ˆ
(R␪ f)(␻)e
j2␲␻( x ⭈ ␪ )
兩␻ 兩 d␻ d␪ (1)
0 ⫺⬁

The above formula, called the filtered back-projection formula, can be im-
plemented in two steps, the filtering step, which in the Fourier domain can
be written as
ˆ
Q̂␪ (␻) = R␪ f(␻)兩␻ 兩 (2)
and the back-projection step,

冕
␲
→ → →
f( x ) = Q␪ ( x ⭈ ␪ ) d␪ (3)
0

Because 兩␻ 兩 is not bounded, and filtering by this filter tends to magnify the
high-frequency noise, it is expedient in practice to multiply this operator by
a smoothing window W(␻) as
ˆ
Q̂␪ (␻) = R␪ f(␻)兩␻ 兩W(␻) (4)
Therefore the reconstruction will result in an approximation of f rather than
f itself. Normally the approximation has the form e * f, where e is an ap-
proximate delta function, called the point spread function [21]. The point
spread function e is related to W(␻) by
W(␻) = ê(␻ cos ␪, ␻ sin ␪)

2.3 Nonlocality of RT Inversion

In Eq. (2), the Radon transform data is filtered by 兩␻ 兩. This operation can
be formulated in the space domain as
Q␪ (t) = H ⭸R␪ f(t)

Copyright 2002 by Marcel Dekker, Inc. All Rights Reserved.

Discovering Diverse Content Through
Random Scribd Documents
nature of the urban tribes, see Varro, L. L. v. 56; Livy i. 43. 13;
Dion. Hal. iv. 14. 1.
[334] Kubitschek, Rom. trib. or. 24 f.; Imp. rom. trib. discr. 2.
[335] Cf. Grotefend, Imp. rom. trib. descr. 7.
[336] Kubitschek, Imp. rom. trib. discr. 2 f.
[337] Cic. Flac. 32. 79 f. On the growth of the tribe, see
Mommsen, Röm. Staatsr. iii. 175 ff.; Kubitschek, ibid. See also
the maps in the latter work.
[338] Flaccus, in Gell. xvii. 7. 5. A list was kept of the estates
comprising a tribe; Cic. ibid.
[339] Cf. the admission of new tribes; Livy vi. 5. 8: “Tribus
quattuor ex novis civibus additae;” viii. 17. 11.
[340] Dion. Hal. iv. 14. 2.
[341] P. 64.
[342] Livy xxix. 37. 3 f.; Soltau, Altröm. Volksversamml. 379, n.
3.
[343] Somewhat different is the view of Mommsen, Röm. Trib. 2
f.; Röm. Forsch. i. 151; Röm. Staatsr. ii. 402; controverted by
Soltau, ibid. 384 ff.
[344] The Romans had but two pursuits, agriculture and war, for
the sedentary occupations were given to slaves and strangers;
Dion. Hal. ii. 28; ix. 25. 2. It was assumed that those who were
without property could take no interest in the state; ibid. iv. 9. 3 f.;
Livy viii. 20. 4.
[345] Cf. Mommsen, Röm. Staatsr. iii. 630.
[346] It is well known too that freedmen were not regularly
employed in military service; Livy x. 21. 4; p. 354 f. below.
[347] Widows and orphans were enrolled in a different list from
that of the tribes, and hence were not included in the statistics of
population which have come down to us; cf. Livy iii. 3. 9; ep. lix;
Plut. Popl. 12; Mommsen, Röm. Staatsr. ii. 365 f., 401. Livy, ii. 56.
3, seems to exclude the clients. Only those lacked membership,
however, who possessed no land. Clients of free birth were as
liable to military service, according to their ratable property, as
any other class of citizens; p. 22.
 [348] Law of the Twelve Tables, in Gell. xvi. 10. 5; Schöll, Leg.
Duod. Tab. Rel. 116; Bruns, Font. iur. 18 f.; Cic. Rosc. Am. 18. 51;
Att. iv. 8 a. 3; Fest. ep. 9; Charis. p. 75 (Keil). The derivation from
ab asse dando proposed by Aelius Stilo, though absurd, was
accepted by Cic. Rep. ii. 22. 40; Top. 2. 10; Fest. ep. 9 (as an
alternative); Isid. Etym. x. 27; Quint. Inst. v. 10. 55. The derivation
ab assidendo is nearer the truth; Vaniček, Griech.-lat. Wörterb.
1012; Lange, Röm. Alt. i. 466; Mommsen, Röm. Staatsr. iii. 237 f.;
Kubitschek, in Pauly-Wissowa, Real-Encycl. i. 426. See also
Varro, De vit. pop. rom. i, in Non. Marc. 67; Gell. xix. 8. 15.
[349] Cic. Rep. ii. 9. 16; 22. 40; P. Nigidius, in Gell. x. 5. 2; Fest.
ep. 9, 119; Pliny, N. H. xviii. 3. 11; Quint. v. 10. 55; Ovid, Fast. v.
281; Vaniček, ibid. 506, 1149.
[350] The army in the field must have consisted largely of men
in patris aut avi potestate, whose names were reported to the
censors, not for taxation but for military service, by those who had
authority over them; cf. Livy xxiv. 11. 7; xliii. 14; Dion. Hal. ix. 36.
3; Fest. ep. 66. Scipio’s complaint (Gell. v. 19. 16: “In alia tribu
patrem, in alia filium suffragium ferre”) indicates that the sons
were regularly enrolled in the tribe of the father. That the list
comprised plebeians only (Niebuhr, Röm. Gesch. i. 457 f.) has
proved untenable; Mommsen, Röm. Forsch. i. 153 f.
[351] Dion. Hal. iv. 14. 2; Livy i. 43. 14; Varro, L. L. v. 181.
[352] Livy, ibid.; Varro, ibid.; cf. p. 63, n. 4 below.
[353] Dion. Hal. iv. 19. 3; Fest. ep. 9; Ennius, in Gell. xvi. 10. 1;
cf. 12 f. Before the introduction of pay for military service in 406
the soldiers bore their own expenses; Livy iv. 59. 11; v. 4. 5; viii. 8.
3; Flor. i. 6. 8; Diod. xiv. 16. 5; Lyd. De mag. i. 45 f.; p. 71 ff.
below.
[354] Plutarch, Cam. 2, makes Camillus the author of the tax on
orphans for the support of the knights’ horses, thus connecting
this measure with the general introduction of pay—a statement of
some importance notwithstanding Kubitschek, in Pauly-Wissowa,
Real-Encycl. i. 683.
[355] Zon. vii. 20: Οἰκόσιτοι ἐστρατεύοντο.
[356] Cic. Rep. v. 2. 3.
[357] Marquardt, Röm. Staatsv. ii. 150 f., 159 f. with citations.
[358] Cic. Rep. ii. 20. 36; Livy i. 43. 9; Plut. Cam. 2.
 [359] Lange, Röm. Alt. i. 469, is of the opinion that before
Servius all the plebeians had this standing, and that Servius left
the newly conquered plebeians in that class, because if admitted
to the army, they might revolt! Cf. Herzog, Röm. Staatsverf. i. 95.
[360] On the meaning of the word, see Pseud. Ascon. 103: “Ut
pro capite suo tributi nomine aera praeberet.” On the removal
from the tribe into this class; Livy iv. 24. 7; xxiv. 18. 6, 8; 43. 3;
xliv. 16. 8. The removal from the tribe is understood when it is not
mentioned; Varro, in Non. Marc. 190; Livy ix. 34. 9; xxvii. 11. 15;
Gell. iv. 12.
[361] Livy vii. 20. 7; Dio Cass. Frag. 33; Strabo v. 2. 3; Gell. xvi.
13. 7; Schol. Hor. Ep. i. 6. 62. On the aerarii and Caerites, see
further Mommsen, Röm. Staatsr. ii. 392-4, 401 ff., 406;
Kubitschek, in Pauly-Wissowa, Real-Encycl. i. 674-6; iii. 1284 f.;
Hülsen, ibid. iii. 1281 f.; see also the works of Herzog, Lange,
Madvig, and Willems.
[362] P. 466, n. 2.
[363] It would be absurd to suppose that while the absolutely
poor citizens could vote in the proletarian century, those who
possessed considerable wealth, though not in land, were
excluded.
 [364] Unutterable confusion was brought into this subject by
Varro, L. L. v. 181: “Tributum dictum a tribubus, quod ea pecunia,
quae populo imperata erat, tributim a singulis pro portione census
exigebatur;” cf. Livy i. 43. 13; Isid. Etym. xvi. 18. 7. Neither is
tributum derived from tribus nor vice versa. Tribuere signifies “to
divide,” “to apportion;” tributum, “that which is apportioned,” tribus
being only indirectly connected with these words; Schlossmann,
in Archiv f. lat. Lexicog. xiv (1905). 25-40.
[365] Livy vi. 14. 12.
[366] Ibid. 32. 1.
[367] Dion. Hal. v. 20; cf. iv. 11. 2; xi. 63. 2; Plut. Popl. 12.
[368] Livy ii. 9. 6; xxiii. 48. 8; xxxiii. 42. 4; xxxix. 7. 5; Pliny, N. H.
xxxiv. 6. 23; Marquardt, Röm. Staatsv. ii. 162, n. 4.
[369] Instances of public expenditure for the equipment or pay
of troops before this date (Dion. Hal. v. 47. 1; viii. 68. 3; ix. 59. 4;
Livy iv. 36. 2) are either exceptional or more probably historical
anticipations of later usage. That before 406 the soldiers drew
pay from their tribes (Mommsen, Röm. Trib. 32; Lange, Röm. Alt.
i. 540) is disproved by Soltau, Altröm Volksversamml. 407 f.
[370] Marquardt, ibid. 164-7.
[371] Cf. Mommsen, Röm. Staatsr. ii. 392.
[372] Varro, L. L. v. 181.
[373] The function of the tribuni aerarii was to pay the soldiers;
Cato, Epist. Quaest. i, in Gell. vi (vii). 10. 2; Varro, v. 181; Fest.
ep. 2; Pliny, N. H. xxxiv. 1. 1. Perhaps they also collected money
into the treasury; Cic. Att. i. 16. 3. From Cato’s statement they
appear to have been financially responsible; and we are informed
that as early as 100 they constituted a rank (ordo) evidently next
below the equites; Cic. Rab. Perd. 9. 27. Under the Aurelian law
of 70 they made up a decury of jurors; Cic. Att. i. 16. 3; Pliny, N.
H. xxxiii. 1. 31. From these facts it is clear that the aerarian
tribunes were officers of the aerarium, but no connection with the
tribes can be discovered; Soltau, Altröm. Volksversamml. 409-12.
[374] Diod. xx. 46; Livy ix. 46. 10 f.; cf. Mommsen, Röm.
Staatsr. ii. 403.
[375] Mommsen, ibid. This class came to an end in the Social
War; Kubitschek, in Pauly-Wissowa, Real-Encycl. iii. 1285.
 [376] In Mommsen’s opinion (Röm. Staatsr. ii. 403) these
censors transferred to the country tribes as many landholding
members of the urban tribes as possible.
[377] Livy ix. 46. 13 f.
[378] Livy xlv. 15.
[379] The expression tribu movere or in aerarios referre was
still used, but meant no more than the transfer from a rural to an
urban tribe and to the aerarian class within the latter; p. 62, n. 7.
[380] Cf. Livy xxiv. 18. 8 f.
[381] Livy xxiv. 43. 2 f.; Cic. Cluent. 42. 120.
[382] P. 86.
[383] I. 43. The account given by Dionysius Hal. iv. 16 f.; vii. 59,
is the same in principle, though slightly different in detail.
[384] P. 52.
[385] Fest. 246. 30; or “discriptio classium,” ibid. 249. 1.
[386] Livy i. 60. 4.
[387] Quoted by Cic. Orat. 46. 156, for the forms “centuria
fabrum” and “procum.” Varro, L. L. vi. 86-8, is an extract from the
Tabulae of later time; cf. Mommsen, Röm. Staatsr. iii. 245, n. 1.
[388] P. 52. Proof of the date is the fact that the ratings are in
the sextantarian as, legally adopted in 269 or 268 (page 86). The
as of this standard was valued at one tenth of a denarius, so that
1000 asses = 100 denarii = 1 mina; Dion. Hal. iv. 16 f.; Polyb. vi.
23. 15: Οἱ ὑπὲρ τὰς μυρίας τιμώμενοι δραχμάς, descriptive of the
highest rating—100,000 asses; Mommsen, Röm. Staatsr. iii. 249,
n. 4; Hill, Greek and Roman Coins, 47. It could not have been
later than 241, in which year the reform of the centuriate
assembly must have been far advanced, if not completed; page
215.
[389] P. 84.
[390] It is wrong to suppose with Soltau, in Jahrb. f. cl. Philol. xli
(1895). 412, n. 6, that all the details of the Servian system were
known only in this way.
[391] Cf. Livy i. 44. 2; Dion. Hal. iv. 15. 1.
[392] Smith, Röm. Timokr. 9 ff., supposes Calpurnius Piso to
have been the intermediary. But a problem in which so many of
the quantities are unknown is incapable of solution.
[393] P. 205, n. 5, 215.
[394] Livy i. 43. 8; Dion. Hal. iv. 18. 2; p. 207.
[395] P. 80.
[396] P. 81.
[397] P. 81.
[398] P. 82 f.
[399] Livy viii. 8. 3; Dion. Hal. iv. 22. 1.
[400] It is unnecessary here to consider the question as to the
historical personality of Servius Tullius. In this volume the name
will be given to the king (or group of kings?) who instituted the so-
called Servian tribes and the military centuries and made a
beginning of the census.
[401] P. 201.
[402] Helbig, Sur les attributes des saliens, in Mémoires de
l’acad. d. inscr. et belles-let. xxxvii (1906). 230 ff.; cf. Comptes
rendus de l’acad. etc. 1904. ii. 206-12. Helbig finds that the
Latino-Etruscan equipments of the time preceding Hellenic
influence, as shown by archaeology, correspond closely with
those of the Salii, whom he regards therefore as religious
survivals from that early civilization. It is from archaeological data,
combined with the well-known equipment of the Salii, that the
close resemblance between the early Latino-Etruscan and the
Mycenaean military system is established.
[403] Not merely the chief, as Helbig, Comptes rendus, 1900.
517, supposes. The ἠνίοχοι καὶ παραβάται who fought at Delium,
and whom he rightly regards as a survival from the age of war-
chariots, acted as a company not as individuals; Diod. xii. 70. 1.
[404] Helbig, Le Currus du roi Romain, in Mélanges Perrot, 167
f. It was like that chiseled on a gravestone found by Dr.
Schliemann on the acropolis of Mycenae, in the main identical
with the Homeric chariot, represented in later time on the famous
sarcophagus at Clazomenae; Pellegrini, in Milani, Studi e
materiali, i. 91-3, 98.
[405] That the army of Romulus—the primitive Roman army—
was a single legion, and that the Servian reform consisted
accordingly in doubling it, is an ancient hypothesis accepted by
some moderns, as Smith, Röm. Timokr. 38 f. An organization in
definite numbers, however, as 1000 from each tribe, cannot arise
till the state has grown sufficiently populous to make up the army
of a part only of its available strength, when folk and army have
ceased to be identical (Schrader, Reallex. 350), and it is agreed
that this condition was not reached till after the adoption of the
Servian reform; Delbrück, Gesch d. Kriegsk. i. 225; Smith, ibid. 52
f., 56.
[406] Il. ii. 362.
[407] Schrader, ibid. For the Sueves, see Caesar, B. G. iv. 1; for
the Lacedaemonian army, see p. 71. The assumption of Helbig,
Comptes rendus, 1904. ii. 209, that the army was composed of
patricians only is altogether unwarranted. Equally groundless is
the notion of Soltau, Altröm. Volksversamml. 250, that the
Homeric army was composed chiefly of nobles with a few light-
armed dependents.
[408] Cf. Liers, Kriegswesen der Alten, 78; Niese in Hist.
Zeitschr. xcviii (1907). 264, 266, 289.
[409] Il. iv. 293 ff.
[410] Represented by the dances of the Salii; Helbig, ibid. 211 f.
[411] Paus. iv. 8. 11; Polyaen. i. 10; Delbrück, Gesch. d.
Kriegsk. i. 30 f.; Niese, in Hist. Zeitschr. xcviii (1907). 274 ff.
[412] Cf. Thuc. v. 70; Polyaen. i. 10.
[413] Cf. Thuc. v. 69. For this and other depths, see Delbrück,
ibid. i. 25; Liers, Kriegswesen der Alten, 45; Lammert, in N. Jahrb.
f. kl. Philol. xiii (1904). 276 f.
[414] Tyrtaeus, Frag. xi (Bergk). For the shield which covered
“hips, legs, breast, and shoulders,” v. 23 f. It was abolished by
Cleomenes III; Plut. Cleom. 11; cf. Liers, ibid. 34; Lammert, ibid.
276 f.
[415] XII. 26; Xen. Anab. i. 2. 16. A public gift of a bronze
cuirass is mentioned by Aristotle, Lac. Pol. 75, Müller, Frag. Hist.
Graec. ii. p. 127. Gilbert, Const. Antiq. 73; Delbrück, ibid. 25,
maintain that the cuirass was a regular part of the equipment.
This is true of soldiers who carried smaller shields.
[416] Beloch, Griech. Gesch. i. 200 f.; cf. Liers, Kriegswesen
der Alten, 34 f.; Droysen, Griech. Kriegsalt. 3 ff.
 [417] Cf. the name of one of these regiments Μεσσοάτης
(Schol. Thuc. iv. 8) derived from the village or local tribe Messoa.
Schol. Aristoph. Lysistr. 453, mentions five by name; cf. Aristotle,
Frag. 541. Perhaps a sixth for guarding the kings was drawn from
all the tribes; Busolt, Griech. Gesch. i. 535 ff. with notes.
Lenschau, in Jahresb. ü. Altwiss. cxxxv. 83, holds that there were
but four phylae.
[418] The name pentecosty indicates that it originally comprised
fifty men, which suggests that the century may have been a
higher group. Before the Peloponnesian War (Thuc. v. 68) the
Lacedaemonian organization had departed far from its original
form.
[419] Droysen, Griech. Kriegsalt. 70; Gilbert, Const. Antiq. 72.
Compulsory service beyond the border ceased with the fortieth
year; Xen. Hell. v. 4. 13.
[420] Cf. Liers, Kriegsw. der Alten, 14.
[421] Busolt, Griech. Gesch. ii. 180 ff.; Helbig, in Mém. de
l’acad. des inscr. xxxvii¹ (1904). 164. But the Athenian army did
not become efficient till long after Solon; cf. Niese, in Hist.
Zeitschr. xcviii (1907). 278-82.
[422] The Romans believed that they got the phalanx from the
Etruscans; Ined. Vat., in Hermes, xxvii (1892). 121 from an early
historian, Fabius Pictor or Posidonius or Polybius (Pais, Anc.
Italy, 323); Diod. xxiii. 2 (Müller); Athen. vi. 106. p. 273 f.;
Wendling, in Hermes, xxviii (1893). 335 ff.; Müller-Deecke,
Etrusker, i. 364 ff.; Smith, Röm. Timokr. 40. The circumstance
does not prove that the Romans were then in subjection to the
Etruscans.
[423] Some of the ancients derive classis from calare, “to call,”
hence “summoning;” Dion. Hal. iv. 18. 2; Quint. Inst. i. 6. 33;
accepted by Walde, Lat. Etym. Wörterb. 125; Soltau, Altröm.
Volksversamml. 242; Lange, Röm. Alt. i. 464. Others connected it
with κᾶλος “firewood,” hence “gathering;” Serv. in Aen. i. 39; Isid.
Etym. xix. 1. 15; Schol. Luc. i. 306. Corssen, Ausspr. i. 494,
proposes to derive it from a root “clat,” which appears in the
Greek κλητεύειν (Lat. *clat-ē-re), Germ. laden, which would still
give the meaning “summoning;” cf. Curtius, Griech. Etym. 139;
Vaniček, Griech. Lat. etym. Wörterb. 143 (*cla-t, cla-t-ti-s).
Mommsen accepted the meaning “summoning” in the early
editions of his History, but rejects it in the Staatsrecht, iii. 262 f.
(cf. his History, English ed. i. 1900. 115 f., 118) on the ground that
however adapted it may have been to the later political classes, it
could not well apply to the fleet and army, and hence could not
belong to the earlier use of the word, which denoted the line in
contrast with those who fought outside the line. But against his
reasoning it could be urged that classis with the idea of
“summoning” first applied to the line of heavy infantry—the only
effective part of the army; and when once the connotation of “line”
had been established, it could easily extend to the fleet.
[424] Gell. vi (vii). 13: “‘Classici’ dicebantur non omnes, qui in
quinque classibus erant, sed primae tantum classis homines, qui
centum et viginti quinque milia aeris ampliusve censi erant. ‘Infra
classem’ autem appellabantur secundae classis ceterarumque
omnium classium, qui minore summa aeris, quod supra dixi,
censebantur. Hoc eo strictim notavi, quoniam in M. Catonis
oratione, qua Voconiam legem suasit, quaeri solet, quid sit
‘classicus,’ quid ‘infra classem;’” Fest. ep. 113; cf. Cic. Verr. II. i.
41. 104; Pseud. Ascon. 188; Gaius ii. 274.
[425] The statement of Diod. xxiii. 2 (Müller), and of the Ined.
Vat. (in Hermes, xxvii. 121) that the Romans derived their round
shield from the Etruscans accords with archaeological evidence
for the use of the round shield by the early Etruscans; Pellegrini,
in Milani, Studi e materiali, i. 91 ff.; Helbig, in Comptes rendus de
l’acad. des inscr. 1904. ii. 196.
[426] The notion of Delbrück, Gesch. d. Kriegsk. i. 227, that the
army was not organized in centuries till after the beginning of the
republic has no foundation whatever.
[427] P. 76. The original number cannot be determined.
[428] Tubero, in Gell. x. 28. 1; Non. Marc. 523. 24. From this
fact it appears that military conditions made a far greater demand
upon the early Romans than upon the Lacedaemonians.
[429] Helbig, in Comptes rendus de l’acad. des inscr. 1900. 516
ff.; Mém. de l’acad. etc. xxxvii¹ (1904). 157 ff.; Hermes xl (1905).
109. The objection of Smith, Röm. Timokr. 37, n. 3, is not well
founded.
[430] Incertus Auctor (Huschke), p. 1.
[431] Ined. Vat., in Hermes xxvii (1892). 121; Helbig, ibid, xl
(1905). 114. The transvectio equitum was instituted in 304; Livy ix.
46. 15. On the close connection of the Roman cavalry with that of
the Greeks of southern Italy, see Pais, Storia di Roma, I. ii. 607, n.
1.
[432] The priores had each two horses; Granius Licinianus xxvi,
p. 29: “Verum de equitibus non omittam, quos Tarquinius ita
constituit, ut priores equites binos equos in proelium ducerent;” cf.
Fest. ep. 221. On the Tarentine cavalry, see Livy xxxiii. 29, 5. The
inference is that the posteriores had one horse each.
[433] Helbig, in Hermes xl (1905). 107. Notizie degli Scavi,
1899. 167, fig. 17 (cf. p. 157); 1900. 325, fig. 28; Pellegrini, in
Milani, Studi e materiali, i. 106.
[434] Pellegrini, ibid. i. 97, fig. 5; 104, fig. 10.
[435] P. 75.
[436] P. 3, n. 8.
[437] VI. 13. 4.
[438] The principal sources are Cic. Rep. ii. 20. 36; 22. 39; Livy
i. 13. 8; 15. 8; 36. 7; 43. 8 f.; Dion. Hal. ii. 13; vi. 13. 4; Pliny, N. H.
xxxiii. (9.) 35; Fest. ep. 55; Plut. Rom. 13. On the basis of these
sources we could reckon an increase to 1800, 3600, or 5400
according to our assumption as to the number of horsemen to the
century; cf. Gerathewohl, Die Reiter und die Rittercenturien, 3-8.
[439] Helbig, in Hermes, xl (1905). 101, 105, 107.
[440] Livy i. 13. 8; Dion. Hal. ii. 13. 1 f.; Fest. ep. 55.
[441] Cic. Rep. ii. 20. 36: Livy i. 36. 2, 7; Fest. 344. 20; ep. 349.
Writers differ slightly in the form of the names.
[442] P. 73, n. 7.
[443] This distinction of rank among the patrician centuries of
the comitia centuriata is proved by the expression “proceres
patricii” in the Censoriae Tabulae, quoted by Fest. 249. 1:
“Procum patricium in descriptione classium, quam fecit Ser.
Tullius, significat procerum. I enim sunt principes;” Cic. Orat. 46.
156: “Centuriam fabrum et procum, ut censoriae tabulae
loquuntur, audeo dicere, non fabrorum aut procorum.” Mommsen,
Röm. Staatsr. iii. 109, n. 1, has rightly referred it to one of the sex
suffragia, for no century outside this group could have been so
designated; cf. Livy ii. 20. 11, who speaks of the cavalry as
proceres iuventutis. The mention of a century of leading patricians
implies the existence of one or more centuries of the less
distinguished members of the same rank, which must have been
the rest of the sex suffragia. The superior rank of the equites in
early Rome is proved by Dion. Hal. ii. 13. 1; iv. 18. 1; Livy i. 43. 8
f.; ii. 20. 11. In ii. 24. 2 Livy implies that the patricians did not
serve on foot (militare), and in iii. 27. 1 he speaks of a patrician
who, as an exception among his rank, served on foot because of
his poverty. In ii. 42 f. he distinguishes the cavalry from the
infantry as patricians from plebeians. The fact that in the political
conflict between the two social classes the patricians often
threatened to carry on foreign wars with the aid merely of their
clients (cf. Dion. Hal. x. 15, 27 f., 43) proves that the phalanx was
essentially plebeian. On the honorable place of the equites in the
camp, see Nitzsch, in Hist. Zeitschr. vii (1862). 145. That the sex
suffragia remained patrician down to the reform of the comitia
centuriata is probable; cf. Sallust, Hist. i. 11, who represents the
struggle between the social classes as continuing to the opening
of the war with Hannibal; see also Mommsen, Röm. Staatsr. iii.
254.
[444] Dion. Hal. ii. 7. 4; cf. Polyb. vi. 25. 1; Varro, L. L. v. 91:
“Turma terima (e in u abiit) quod ter deni equites ex tribus
tribubus Titiensium Ramnium Lucerum fiebant: itaque primi
singularum decuriones dicti, qui ab eo in singulis turmis sunt
etiamnunc terni;” cf. Curiatius, in Fest. 355. 6.
[445] Cf. Polyb. vi. 25. 1.
[446] Three hundred is given as normal by Polyb. i. 16. 2; vi.
20. 9. In iii. 107. 10 f. he states it at 200, increased to 300 when to
meet extraordinary cases the legion was strengthened to 5000;
cf. ii. 24. 3. Livy, xxii. 36. 3, agrees with the latter statement.
Mommsen, Röm. Staatsr. iii. 477, believes that the normal
number was 300, decreased to 200 when a greater number of
legions was levied.
[447] Niese, Hist. Zeitschr. xcviii (1907). 283, rightly assumes
that the first and second classes at Athens were not cavalry;
Helbig is right in understanding them to be mounted hoplites.
Niese’s criticism (ibid. 287 and n. 1) of Helbig’s view is not
convincing.
[448] Considerable time was required for the establishment of
the earliest known meaning of classis before the second and third
divisions were added.
[449] This is a conjecture of Bruncke, in Philol. xl (1881). 362,
favored by Delbrück, Gesch. d. Kriegsk. i. 222.
 [450] P. 79, 86.
[451] Usually scholars (cf. Domazewski, in Pauly-Wissowa,
Real-Encycl. iii. 1953 f.; Delbrück, Gesch. d. Kriegsk. i. 227;
Smith, Röm. Timokr. 39) assume fifteen centuries for the fifth
rating, on the authority of Livy i. 43. 7; Dion. Hal. iv. 17. 2; vii. 59.
5. But our knowledge of the phalanx is only inference, which to be
acceptable must have at least the merit of possibility. The number
fifteen is wrong because it could not have been divided evenly
between the two legions; and on the other hand it will be shown
later (p. 208) that in all probability the fifteenth century was not
military but was added in the make up of the comitia centuriata.
[452] Müller, in Philol. xxxiv (1876). 129, is right in supposing
that the legion was strengthened between the time of Servius and
387, but it was not in the way he assumes. The tradition of a
legion (half phalanx) of 4000 men is preserved in Livy vi. 22. 8.
[453] Polyb. vi. 20.
[454] Cf. Smith, Röm. Timokr. 121 ff.
[455] Livy iv. 46. 1: “Dilectum haberi non ex toto passim populo
placuit: decem tribus sorte ductae sunt. Ex his scriptos iuniores
duo tribuni ad bellum duxere.” If this passage does not state a
historical fact, at least it gives the idea of the writer as to the
custom of earlier time.
[456] P. 72, 76.
[457] Cf. Smith, Röm. Timokr. 51 ff.
[458] In time of especial danger, however, the legion was
increased to five thousand; Polyb. vi. 20. 8.
[459] Cf. Mommsen, Röm. Staatsr. iii. 268, n. 2.
[460] That the phalanx was a comparatively late institution at
Rome, or that it was slow in becoming the only military system, is
indicated by the survival in tradition of a more primitive mode of
warfare. Sometimes in the early republic a single gens with its
clients took the field; for the Fabian gens, see Livy ii. 48 ff. Often
the patricians threatened to arm their clients, to carry on a war
without the aid of the troublesome plebeians; cf. Dion. Hal. x. 15,
27 f., 43. As there was no motive in later time for the invention of
such stories, they must contain a kernel of real tradition; hence
they could not go back to the sixth century, and it is difficult to
believe that they are so old as the fifth.
 Collateral evidence that the second and third divisions were
instituted relatively late may be found in the circumstance that the
scutum, the distinctive piece of armor of these divisions, was
introduced no earlier than the age of Camillus—the period of the
war with Veii and the Gallic conflagration; Livy viii. 8. 3; Müller-
Deecke, Etrusker, i. 366. It was Samnite (Athen. vi. 106, p. 273 f.;
cf. Sall. Cat. 51), and was therefore probably adopted in the
fourth century when Rome first came into contact with that
people.
[461] It is evident to the reader that these proportions are those
of the discriptio centuriarum of Livy and Dionysius (p. 66 above),
and it will be made clear below (p. 86) that the ratings were
originally in terms of iugera, the minima of the five ratings being in
all probability 20, 15, 10, 5, and 2½ or 2 iugera respectively.
[462] For the date, see Mommsen, Röm. Staatsr. ii. 334 f.;
Kubitschek, in Pauly-Wissowa, Real-Encycl. iii. 1902 f.; Pais,
Storia di Roma, I. ii. 13, 33 f.
[463] There may be some truth in the etymology suggested by
Varro, L. L. v. 89; cf. Soltau, Altröm. Volksversamml. 256.
[464] Cf. Liers, Kriegsw. d. Alten, 46.
[465] Dionysius Hal. iv. 17. 1, includes the fourth rating in the
phalanx of heavy infantry. For other possibilities of arrangement,
see Smith, Röm. Timokr. 46 f.
[466] Thuc. v. 68; p. 86 above.
[467] Delbrück, Gesch. d. Kriegsk. i. 229; Smith, Röm. Timokr.
45 ff. That the second and third divisions of the phalanx were
sometimes withdrawn to operate on the flanks (Soltau, Altröm.
Volksversamml. 249) is possible, though we have no proof of it.
[468] P. 76. From early times the Greek and Italian states kept
arsenals with which to arm the poor in crises; Liers, Kriegsw. d.
Alten, 36 f.
[469] P. 84.
[470] Fest. ep. 14, 18, 369; Varro, L. L. vii. 56-58. From them
the centurions and decurions engaged their servants; Cato, in
Varro, L. L. vii. 58; Varro, Vit. pop. rom. iii, in Non. Marc. 520;
Veget. ii. 19. Hence they served the civil magistrates as
attendants; cf. Censoriae Tabulae, in Varro, L. L. vi. 88; Livy iii.
33. 8; Suet. Caes. 20; Non. Marc. 59. They must have
corresponded with the squires of the Greek and Roman cavalry;
p. 73. They were sometimes called adscriptivi, or as carriers
ferentarii. If, as has been suggested, the secretaries and other
attendants of the higher officers were also drawn from them, this
circumstance would help explain the honor attaching to the
collegium accensorum velatorum of imperial time; Mommsen,
Röm. Staatsr. iii. 289; Delbrück, Gesch. d. Kriegsk. i. 233.
[471] Notwithstanding Kubitschek, in Pauly-Wissowa, Real-
Encycl. i. 135 f.
[472] Livy viii. 8. 8. Leinveber, in Philol. N. F. xv (1902). 36,
estimates 558 accensi to the legion.
[473] The cornicines tubicinesque; Livy i. 43. 7.
[474] The cornicines marched in front of the banners; Joseph.
Bell. Iud. v. 48; Fiebiger, in Pauly-Wissowa, Real-Encycl. iv. 1602.
[475] The number is unknown. In the legio III Augusta there
were thirty-six cornicines; CIL. vii. 2557; Fiebiger, ibid. 1603.
[476] Livy i. 43. 3.
[477] Varro, L. L. v. 88: “Centuria qui sub uno centurione sunt,
quorum centenarius iustus numerus;” Fest. ep. 53: “Centuria ...
significat ... in re militari centum homines;” Isid. Etym. ix. 3. 48; cf.
Huschke, Verf. d. Serv. 107.
[478] Estimates have been made by Müller, in Philol. xxxiv
(1876). 127; Delbrück, Gesch. d. Kriegsk. i. 224; Beloch, Bevölk.
d. griech.-röm. Welt, 42 f.; Smith, Röm. Timokr. 67. In the United
States the ratio is more than four to one; Special Reports: Suppl.
Analysis and Derivative Tables, Twelfth Census of the United
States, 1900, Washington, 1906. p. 170 f. The estimate given in
the text is based upon the “Deutsche Sterbetafel” for men, in E.
Czuber, Warscheinlichkeitsrechnung (Leipzig, 1903), p. 572, 574.
The ratio is almost exactly three.
[479] Livy i. 43. 2. For the year 401, see Livy v. 10. 4: “Nec
iuniores modo conscripti, sed seniores etiam coacti nomina dare,
ut urbis custodiam agerent;” for 389, vi. 2. 6; for 386, vi. 6. 14; for
296, x. 21. 4: “Nec ingenui modo aut iuniores sacramento adacti,
sed seniorum etiam cohortes factae libertinique centuriati. Et
defendendae urbis consilia agitabantur;” cf. Mommsen, Röm.
Staatsr. ii. 409, n. 5. The last of the definite instances here
mentioned could alone be historical, and in this case not
centuriae or legiones but cohortes seniorum are spoken of.
[480] Cf. Delbrück, Gesch. d. Kriegsk. i. 227 f.
 [481] If the senior centuries were formed in the way assumed
by Mommsen, Röm. Staatsr. iii. 261 (“Nicht selbständig gebildet
worden, sondern daraus hervorgegangen, dass wer aus einer
Centurie des ersten Aufgebots Alters halber ausschied, damit in
die entsprechende Centurie des zweiten Aufgebots eintrat”),
about a half generation must have been required to evolve them.
An objection to his idea is that the military centuries as well as the
legions were formed anew at each year’s levy (Polyb. vi. 20, 24),
whereas the political centuries were made up by the censors (cf.
Cic. Rep. ii. 22. 40: “In una centuria censebantur”), doubtless
modified annually by the consuls. A military century and a political
century accordingly could not have been composed of the same
men.
The Tabulae Iuniorum contained the names of all juniors in
honorable service in the field; Livy xxiv. 18. 7. Tabulae Seniorum
are not mentioned. Classis Iuniorum (Fest. 246. 30) may apply to
all eighty-five (or eighty-four) centuries of juniors, as Lange, Röm.
Alt. i. 474, supposes, or to the first class; Tubero, Historiae, i, in
Gell. x. 28. 1: “Scripsit Servium Tullium regem, populi Romani
cum illas quinque classes iuniorum census faciendi gratia
institueret.” It is doubtful whether there was a separate list of
seniors.
[482] Cic. Rep. ii. 22. 40: “Illarum autem sex et nonaginta
centuriarum in una centuria tum quidem plures censebantur quam
paene in prima classe tota.”
[483] Soltau, Altröm. Volksversamml. 240.
[484] The confusion of the comitia with the army, which the
ancient writers began, the moderns have intensified till the subject
has become utterly incomprehensible. Chiefly to Genz,
Servianische Centurienverfassung (1874) and Soltau, Alröm.
Volksversammlungen (1880) belongs the credit of putting in a
clear light the fact that the original Servian organization was an
army. Both authors, however, have made the fundamental
mistake of supposing that for a time during the early republic the
army officiated as an assembly.
[485] Livy xxiv. 8. 19.
[486] After the inclusion of the Tribus Clustumina; Beloch, Ital.
Bund, 74; Smith, Röm. Timokr. 58, n. 1.
[487] Delbrück, Gesch. d. Kriegsk. i. 223 f.; Smith, Röm.
Timokr. 58.
 [488] Beloch, Bevölk. d. griech.-röm. Welt, 53; Meyer, Forsch.
z. alt. Gesch. ii. 162, n. 3; Delbrück, ibid. i. 14. Ferrero’s estimate
(Greatness and Decline of Rome, i. 1) of a total population of
150,000 seems to be too large.
[489] P. 81.
[490] Cf. Liers, Kriegsw. d. Alten, 10.
[491] Ascribed to Camillus; Plut. Cam. 40; cf. Fröhlich, Gesch.
d. Kriegsführung und Kriegskunst der Römer zur Zeit der Rep.;
Schiller, Röm. Alt. 708.
[492] P. 80; cf. 63.
[493] Fröhlich, ibid. 21 f.; Schiller, ibid.
[494] P. 76.
[495] Fest. 189. 13; ep. 56, 225; Fabius Pictor, Annales, i, in
Gell. x. 15. 3 f.
[496] Gell. i. 11. 3; Vergil, Aen. vii. 716: “Hortinae classes.”
[497] Gell. vi (vii). 13. 3: “In M. Catonis oratione, qua Voconiam
legem suasit, quaeri solet, quid sit classicus, quid infra classem;”
p. 90 below.
[498] CIL. i. 200 (Lex Agr.). 37: (“Recuperatores ex ci)vibus L
quei classis primae sient, XI dato.”
[499] P. 66 f.; cf. Fest. 249. 1: “In descriptione classium quam
fecit Ser. Tullius.” The attempt of Smith, Röm. Timokr., especially
140 ff., to prove that the five classes were introduced by the
censors of 179 has nothing in its favor. It rests upon Livy xl. 51. 9:
“Mutarunt suffragia, regionatimque generibus hominum causisque
et quaestibus tribus descripserunt.” This passage makes no
reference to the classes. In “generibus hominum” are included
chiefly the “genus ingenuum” and the “genus libertinum.” “Causis”
applies to those conditions of the libertini, such as the possession
of children of a definite age, which might serve as a ground for
enrolment in a rural tribe; and “quaestibus” refers to the distinction
between landowners and the “opifices et sellularii” of the city.
“They changed the arrangement for voting, and drew up the tribal
lists on a local basis according to the social orders, the
conditions, and the callings of men;” cf. Lange, Röm. Alt. ii. 265 f.;
p. 354 f. below. Among the many objections to Smith’s theory
these two may be mentioned: if the classes were introduced at
this late historical time, (1) they would not have been ascribed to
Servius Tullius; (2) they would have been adapted to the
economic conditions of the second century b.c., whereas in 179
they were largely outgrown by the depreciation of the standard of
value, the increase in the cost of living, and the growth of
enormous estates. The Römische Timokratie is ably written, but
its main thesis—the institution of the classes in the second
century b.c.—remains unproved.
[500] P. 64.
[501] Verf. d. Serv. 643 f. et passim. He made a mistake
however in supposing that from the beginning land was valued in
terms of money.
[502] Mommsen, Röm. Trib. 111; Röm. Staatsr. iii. 247 ff.;
Kübler, in Pauly-Wissowa, Real-Encycl. iii. 2631. When the
change was made from a land to a money rating, the land of the
fifth class was appraised relatively higher than that of the others.
Neumann, Grundherrsch. d. röm. Rep. 9 f., prefers to assume 16
(= 2 + 14) iugera for the highest class in order to explain the often
mentioned estates of seven and fourteen iugera. But it is difficult
to work out a consistent scheme on this basis. Smith, Röm.
Timokr. 78 ff. et passim, strongly objects to the view in any form,
as he doubts the existence of the Servian classes. In general he
has greatly exaggerated the difficulties of their administration.
[503] Sall. Iug. 86; Gell. xvi. 10. 14, 16; cf. Cass. Hem. 21
(Peter, Reliquiae, i. 102 f.).
[504] Haeberlin, in Riv. ital. numis. xix (1906). 614 f.
[505] Samwer-Bahrfeldt, Gesch. d. alt. röm. Münzw. 176 f.; Hill,
Greek and Roman Coins, 47, 49, n. 1; Kubitschek, in Pauly-
Wissowa, Real-Encycl. ii. 1509 ff.; Hultsch, ibid. v. 206; Regling,
in Klio, vi (1906). 503. Babelon, Trait. d. mon. Grecq. et Rom. i.
595, still holds the view that the triental as was introduced in 269;
cf. his Orig. d. la mon. 376; Mon. d. la rép. Rom. i. 37.
[506] P. 66 f.
[507] As silver is at present worth 51¼ cents an ounce (so
quoted in New York, Sept. 5, 1908), a denarius (= ⅟₇₂ lb. Troy) of
the coinage preceding 217 is worth by weight today 8½ cents. A
more just comparison would be based on the present coined
values. As a dollar contains 371¼ grains of silver, a denarius
would be worth 21½ cents; or with a liberal allowance for the
alloy, we might say about 20 cents. The sesterce, ¼ denarius,
would therefore be equivalent to five cents. An estate of 100,000
asses of heavy weight (sesterces) would be worth about $5000,
of the sextantarian standard $2000. It is hardly possible that so
large a proportion of the population as was contained in the first
class should average the former amount of wealth to the family. In
fact the purchasing power of money was enormously higher than
these equivalents indicate. In 430 the value of an ox or cow was
legally set at 100 libral asses and of a sheep at ten. Reckoning a
beef at the low modern value of $45, and a sheep at $4.50, we
obtain a value of 45 cents for the libral as, or 22½ cents for one of
5 oz. weight (sesterce), which would give the denarius a
purchasing power of 90 cents.
[508] Röm. Staatsr. iii. 249. In his History (Eng. ed. 1900), iii.
50, he expresses some doubt as to the numbers.
[509] I. 43; cf. p. 66.
[510] IV. 17. 2.
[511] Plut. Popl. 21.
[512] The view of Goguet, Centuries, 29 (following Niebuhr),
that Livy has made a mistake, is not so likely.
[513] VI. 19. 2: (All must serve in war) πλὴν τῶν ὑπὸ τὰς
τετρακοσίας δραχμὰς τετιμημένων· τούτους δὲ παριᾶσι πάντας εἰς
τὴν ναυτικήν. That it was the minimal rating of the fifth class, and
not a still lower rating for military use only, is proved by a
statement of Sall. Iug. 86, that till the time of Marius the soldiers
were drawn from the classes.
[514] Cf. Mommsen, Röm. Staatsr. iii. 251.
[515] Commercially the denarius was then, after 217, worth
sixteen asses; Hultsch, in Pauly-Wissowa, Real-Encycl. v. 209.
[516] Cic. Rep. ii. 22. 40; Gell. xvi. 10. 10.
[517] XVI. 10. 10.
[518] Cf. Kübler, in Pauly-Wissowa, Real-Encycl. iii. 1522.
[519] This interpretation differs slightly from that of Mommsen,
Röm. Staatsr. iii. 237.
[520] In like manner those possessing above 100,000 asses
were at times divided into groups for the distribution of military
burdens according to wealth; cf. Livy xxiv. II. 7-9. This too has no
reference to the organization of the comitia.
 [521] N. H. xxxiii. 3. 43: “Maximus census C̅ X̅ assium fuit illo
(Servio) rege, et ideo haec prima classis.”
[522] Fest. ep. 113.
[523] VI (VII). 13.
[524] Plut. Popl. 21; Huschke, Verf. d. Serv. 164.
[525] VI. 23. 15.
[526] I. 43. 2.
[527] IV. 16. 2.
[528] After the adoption of the as of an ounce weight in 217,
sixteen asses of this standard were considered equivalent to a
denarius or a drachma, which would give a rating of 160,000
asses for those who wore the cuirass. But the military pay was
still reckoned at ten asses to the denarius (Pliny, N. H. xxxiii. 3.
45); the censors seem to have used the same ratio (Livy xxxix.
44. 2 f. compared with Plut. Cat. Mai. 18); and it is therefore
highly probable that in this statement Polybius intended to
express in drachmas the value of 100,000 asses. Taken in its
entirety, the passage sufficiently proves that reference is to the
highest class; the majority (οἱ πολλοί) of soldiers, he says, have
breastplates, but those rated above 10,000 drachmas wear
cuirasses. If, as Belot, Rév. écon. et mon. 77 ff., imagines, the
sum of 100,000 asses fell below the rating of the lowest class,
there would hardly have been a soldier without the cuirass.
[529] Gaius ii. 274. That registration was necessary is proved
by Cic. Verr. II. i. 41. 104 ff. By the word “censi” Cicero does not
mean to designate any group or division of citizens; he simply
refers to the fact of registration. P. Annius Asellus, of whom he
speaks, had not been registered, or in any case at that sum, and
hence was not technically liable to the law; but the value of his
estate could be ascertained by authority of a court of justice,
according to Greenidge, Leg. Proced. 95 f. Mommsen held the
opinion, on the contrary (Abhdl. d. Akad. d. Wiss. zu Berlin, 1863.
468 f.), that the incensi were absolutely free from the law.
[530] P. 85 above.
[531] VI (VII). 13. For his rating of 125,000 asses for the first
class, see p. 89.
[532] N. 5 above.
[533] Dio Cass. lvi. 10. 2; Pseud. Ascon. 188.
 [534] Cf. Mommsen, Röm. Staatsr. iii. 249, n. 4; Greenidge,
Leg. Proced. 95.
[535] The part containing this reference was not essentially
later than the enactment of the Voconian law (p. 361).
[536] P. 403.
[537] XLV. 15. 2.
[538] Röm. Staatsr. iii. 249, n. 2.
[539] P. 90, n. 4.
[540] First offered in his Histoire des chevaliers, i (Paris, 1866),
and afterward defended in his Révolution économique et
monétaire ... à Rome (1885).
[541] Cf. Rév. écon. et mon. 82.
[542] Livy xxiv. 11. 7 f.
[543] Ibid. § 5.
[544] Marquardt, Röm. Staatsv. ii. 498 f.
[545] Rév. écon. et mon. 50. The Roman and Campanian (cives
sine suffragio) knights together amounted to 23,000; Polyb. ii. 24.
14.
[546] About 270,000 in 220; Livy ep. xx.
[547] Even with this understanding we shall have to assume for
the requisition of 214 a division between 100,000 and 300,000—
those rated at 100,000-200,000 asses furnishing two and those at
200,000-300,000 asses three sailors. Otherwise the number of
sailors will be greatly in excess of the need.
[548] Similar conditions exist at present in America. The
monstrous luxury of the few and the heavy fines recently imposed
on the Standard Oil Company do not prove all Americans to be
wealthy.
[549] P. 61 f.
[550] Livy i. 43. 9; Cic. Rep. ii. 20. 36; Fest. ep. 81, 221; Gaius
iv. 27.
[551] Gaius iv. 27.
[552] Rep. ii. 20. 36.
[553] I. 43. 9.