tomography with medipix2 semiconductor pixel detector
TRANSCRIPT
Tomography with Medipix2Semiconductor Pixel Detector
by
Carlos Felipe Roa Garcıa
A thesis submitted in partial fulfillment for theDegree of Physicist
Director
Dr. Bernardo Gomez Moreno
Universidad de los AndesFacultad de Ciencias
Departamento de FısicaBogota D.C., Colombia
January 2009
“A man is not old unitl regrets take place of dreams”
John Barrymore
UNIVERSIDAD DE LOS ANDES
AbstractFacultad de Ciencias
Departamento de Fısica
Tomography with Medipix2 Semiconductor Pixel Detector
by Carlos Felipe Roa Garcıa
The Medipix2 pixel presents features that make it interesting for medical applications
that demand high resolution imaging. An overview of biomedical imaging is presented
with a discussion on current developments, including the Medipix2 detector. The mea-
surements done to characterize and calibrate the Medipix2 are presented. The energy
calibration of the detector and its capability of setting an energy window are used to
take radiographies at different energies. A set of projections is obtained for two phan-
toms and an object to perform their tomographical reconstruction. Some slices of the
reconstructed samples are shown along with a surface render of each object.
Acknowledgements
Agradezco a todas las personas que contribuyeron de alguna manera a la realizacion de
este proyecto. A mis padres y hermanos por el apoyo incondicional y por estar junto a
mı siempre y en especial cuando mas los he necesitado. Al Profesor Bernardo Gomez
Moreno por su colaboracion y su sabidurıa. Muchas gracias por sus ensenanzas y por
darme la oportunidad de trabajar en este tema. A los Profesores Carlos Avila y Juan
Carlos Sanabria por estar pendientes del avance del proyecto y por sus ideas valiosas a lo
largo del proceso. Al Ing. Marco Antonio Gonzalez y a Luis Carlos Gomez por su valiosa
ayuda en la construccion del montaje. A la gente de fısica por los momentos compartidos.
Agradezco tambien a todo aquel que lea esto y crea que deba ser mencionado.
I would like to acknowledge the following organizations that have contributed to the
development of this project:
• The Medipix Collaboration, specially Dr. Michael Campbell, Dr. Xavier Llopart
and Dr. Lukas Tlustos.
• The Institute of Experimental and Applied Physics (IEAP) of the Czech Technical
University in Prague. In particular, Dr. Carlos Granja.
• The School of Science of The Universidad de los Andes.
• The Department of Physics of The Universidad de los Andes.
• The High Energy Physics Group of The Universidad de los Andes.
• The High Energy Physics Latinamerican-European Network (HELEN).
• The Colombian Institute for the Development of Science and Technology (COL-
CIENCIAS).
I gratefully acknowledge the developers of the Octopus tomography reconstruction pack-
age, from the Ghent University, who made available a fully-functional version of the Oc-
topus software for this project. In particular, I would like to thank Dr. Manuel Dierick
for his support during the realization of this project.
iii
Contents
Abstract ii
Acknowledgements iii
List of Figures vi
List of Tables viii
1 Introduction to Biomedical Imaging 11.1 Principles of Radioactivity . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.2 X-ray Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41.3 Measurement of Radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . 61.4 Detection of Radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4.1 Scintillation Detectors . . . . . . . . . . . . . . . . . . . . . . . . . 81.4.2 Charge-Coupled Devices . . . . . . . . . . . . . . . . . . . . . . . . 81.4.3 Fast CT Scanners . . . . . . . . . . . . . . . . . . . . . . . . . . . 91.4.4 Energy-Resolving Methods . . . . . . . . . . . . . . . . . . . . . . 101.4.5 Single-Photon Detection . . . . . . . . . . . . . . . . . . . . . . . . 10
2 The Medipix2 Detector 122.1 The Medipix2 Chip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.1.1 The Analog Part . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132.1.2 The Digital Part . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.1.3 The Chip Periphery . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 The Mpix2MXR20 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162.3 The USB Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.4 The Pixelman Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3 Preliminary Measurements and Energy Calibration 213.1 Threshold Equalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213.2 DACs Scan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233.3 Energy Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253.4 Validation of Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4 Radiographical Imaging with Medipix2 31
iv
Contents v
4.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314.2 Flat Field Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.2.1 Foils Phantom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334.2.2 Capacitor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3 Radiographies at Different Energies . . . . . . . . . . . . . . . . . . . . . . 354.4 Beam Hardening Correction . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.4.1 Foils Phantom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384.4.2 Capacitor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5 Methods for Tomographical Reconstruction 415.1 The Filtered Back Projection Method . . . . . . . . . . . . . . . . . . . . 41
5.1.1 The Projection of an Object . . . . . . . . . . . . . . . . . . . . . . 415.1.2 The Fourier Slice Theorem . . . . . . . . . . . . . . . . . . . . . . 425.1.3 The Filtered Back Projection . . . . . . . . . . . . . . . . . . . . . 44
5.2 Iterative Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
6 Tomographical Imaging with Medipix2 476.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476.2 Tomographical Reconstruction with Octopus 8.3 . . . . . . . . . . . . . . 48
6.2.1 Three Objects Phantom . . . . . . . . . . . . . . . . . . . . . . . . 496.2.2 Helicoidal Wire Phantom . . . . . . . . . . . . . . . . . . . . . . . 506.2.3 Cone Shell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7 Final Discussion 54
A Specifications of Step Motor and Driver 55
B External Control of the Step Motor Driver 58
C The PHYWE X-ray Unit 60
D The Medipix2 Support and the Sample Holder 62
E Comments on Octopus 8.3 64
Bibliography 65
List of Figures
1.1 Radiography of Ms. Rontgen’s Hand . . . . . . . . . . . . . . . . . . . . . 21.2 PET Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41.3 X-ray Tube Schematic and Spectrum . . . . . . . . . . . . . . . . . . . . . 51.4 Schematic of a CCD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91.5 Largest Commercial CCD . . . . . . . . . . . . . . . . . . . . . . . . . . . 91.6 Example of an ERM for Radiographic Imaging . . . . . . . . . . . . . . . 10
2.1 Medipix2 Chip Plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132.2 Medipix2 Pixel Schematic and Layout . . . . . . . . . . . . . . . . . . . . 142.3 Mpix2MXR20 Schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172.4 The USB Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.5 The Pixelman Principal Window . . . . . . . . . . . . . . . . . . . . . . . 192.6 The Pixelman Preview Window . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1 THL Equalization Distribution . . . . . . . . . . . . . . . . . . . . . . . . 223.2 THH Equalization Distribution . . . . . . . . . . . . . . . . . . . . . . . . 233.3 THL DAC Scan with Test Pulses . . . . . . . . . . . . . . . . . . . . . . . 243.4 Copper and Molybdenum X-ray Tube Spectra . . . . . . . . . . . . . . . . 263.5 Derivative of the THL DAC Scan for Cu and Mo Tubes . . . . . . . . . . 273.6 Cross Sections for Ni, Cu and Zn . . . . . . . . . . . . . . . . . . . . . . . 283.7 Derivative of the THL DAC Scan for Ni, Cu and Zn Foils . . . . . . . . . 293.8 Energy Calibration and Validation . . . . . . . . . . . . . . . . . . . . . . 30
4.1 Medipix2 Inside the PHYWE X-ray Unit . . . . . . . . . . . . . . . . . . 324.2 Comparison of Raw and FF Corrected Image of Foils Phantom . . . . . . 334.3 Photograph of Capacitor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344.4 Comparison of Raw and FF Corrected Image of Capacitor . . . . . . . . . 354.5 Radiography of the Capacitor at Different Energies . . . . . . . . . . . . . 364.6 Comparison of Raw and STC-Corrected Image of Foils Phantom . . . . . 384.7 Comparison of Raw and STC-Corrected Image of Capacitor . . . . . . . . 39
5.1 Geometry Definition for Tomographical Reconstruction . . . . . . . . . . 425.2 Comparison Between FBP and OSEM Reconstructions . . . . . . . . . . . 46
6.1 Sinogram Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486.2 Photograph and Projection of the Three Objects Phantom . . . . . . . . . 496.3 Reconstructed Slice and Multi-surface Render of Three Objects Phantom 506.4 Photograph and Projection of the Helicoidal Wire Phantom . . . . . . . . 506.5 Reconstructed Slice and Multi-surface Render of Helicoidal Wire Phantom 51
vi
List of Figures vii
6.6 Photograph and Projections of the Cone Shell . . . . . . . . . . . . . . . . 526.7 Reconstructed Slices and Multi-surface Renders of Cone Shell . . . . . . . 53
A.1 Photograph of the Step Motor of Sample Holder . . . . . . . . . . . . . . 55A.2 Dimensions of Step Motor . . . . . . . . . . . . . . . . . . . . . . . . . . . 56A.3 Dimensions of Step Motor Driver . . . . . . . . . . . . . . . . . . . . . . . 57
B.1 Interface of the LabVIEW Program . . . . . . . . . . . . . . . . . . . . . . 58B.2 LabVIEW Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
C.1 PHYWE X-ray Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
D.1 Medipix2 Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62D.2 Sample Holder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63D.3 Experimental Setup For Tomography Imaging . . . . . . . . . . . . . . . . 63
List of Tables
1.1 Weighting Factors for Different Tissues . . . . . . . . . . . . . . . . . . . . 7
3.1 Characteristic Lines for Cu and Mo X-ray Tubes . . . . . . . . . . . . . . 253.2 Values from Energy Calibration with Cu and Mo X-ray Tubes . . . . . . . 283.3 Values from Energy Calibration with Ni, Cu and Zn Foils . . . . . . . . . 29
viii
Dedicated to my parents and siblings
ix
Chapter 1
Introduction to Biomedical
Imaging
The subject of Biomedical Imaging arises from the requirement of studying the internal
body organs and tissues with non-destructive techniques that provide accurate informa-
tion of the condition of a patient. Since the early ages of medicine until the twentieth
century, there were no means of having a clear image of internal organs without looking
directly at them. The development of instruments capable of taking such images was
based on rather simple physical concepts and improved with the power of computer
technology. Today, it is possible to obtain very accurate images of the bones, organs,
vessels and even very small regions of interest, to help in the diagnosis and treatment of
diseases.
In 1895, Wilhelm Conrad Rontgen, discovered, almost by accident, a new kind of ra-
diation that could penetrate solid objects and caused the darkening of a photographic
plate. With his discovery, which he called X-rays, on December 22, 1895, he was able to
take a radiography of his wife’s hand, in which the bones could be easily identified (see
Figure 1.1).[1] Rontgen’s discovery soon became one of the most important advances in
Medicine, providing it allowed to study the body without opening it.
Another important advance by Allan Mcleod Cormack and Godfrey Newbold Hounsfield
complemented the discovery of X-rays and revolutionized the medical practice. In 1963,
Cormack demonstrated mathematically that it is possible to reconstruct a slice of an
image by its projections.[2, 3] Some years after, Hounsfield invented a machine that
could produce digital axial images with medical application. Cormack and Hounsfield
were awarded, in 1979, the Nobel prize in Medicine for their work.[4]
1
Chapter 1. Introduction to Biomedical Imaging 2
Figure 1.1: The first radiography taken by W. K. Rontgen of his wife’s hand.[1]
The field of Biomedical Imaging has been an interdisciplinary collaboration between
Physics and Medicine. It is founded on the principle of existence and characteristics
of X-rays and has been extended to other types of radiation, such as gamma rays and
positrons. It is possible to obtain images of the body of a patient by (1) the emission
of radiation from a source inside the body and (2) the measurement of the radiation
that passes through the body. In the first mechanism, a radioactive source or tracer
is administered to the patient and its radiation is measured. This gives metabolical
and anatomical information, because the tracer is carried by a substance that some
organs use and the measurement of radiation can give information on the position of
the source. The second mechanism is based on the absorption of radiation by different
tissues, leading to the visualization of the anatomy of the body, in a similar way a
shadow created by an object gives information of its geometry. Both approaches have
required mutual cooperation between physiologists and physicists. The extension of
Rontgen’s discovery and the advances of Cormack and Hounsfield to other types of
radiation required Becquerel’s, and Marie and Pierre Curie’s knowledge of radioactivity,
as well as the work of J. J. Thomson, Rutherford and Bohr on the atomic model.
Engineers have been important contributors to this field, because it was important to get
knowledge in computer technology and data processing techniques, in order to produce
commercial machines that could take useful images of the body.[4]
Chapter 1. Introduction to Biomedical Imaging 3
1.1 Principles of Radioactivity
Radioactivity is the phenomenon ocurring inside an atomic nucleus, in which, an atom
that has become unstable releases some of its energy by emitting a particle or electro-
magnetic wave. This causes the nucleus to suffer a transformation into a more stable
form or to continue emitting particles until it becomes stable. The most common proc-
cesses by which the emission of particles takes place are the alpha, beta and gamma
decays.
• Alpha Decay. An alpha particle, composed by two protons and two neutrons,
which resemble a helium nucleus is produced (42He). It happens in the atoms
of uranium and radium and is sometimes accompanied by a gamma decay. An
example of this reaction is:
23892 U →234
90 Th + α (1.1)
The number of protons and neutrons must be conserved, so the difference in atomic
number between the original and the produced atom must equal to two.
• Beta Decay. Involves the production of electrons (β−) or positrons (β+) in the
nucleus of an atom. Despite the nucleus has protons and neutrons, it can produce
an electron by the transformation of a neutron into a proton1. An example of a
reaction in which beta decay takes place is:
146 C →14
7 N + e− + ν (1.2)
The atomic number increases as a neutron is transformed into a proton, but the
overall number of nuleons remains constant. The reverse reaction is also possible,
but in this case, the decay produces a positron and a neutrino (the antiparticles2
of the electron and the antineutrino, respectively).
127 N →12
6 C + e+ + ν (1.3)
The atomic number decreases as a proton is transformed into a neutron, leaving
the total number of nucleons constant.[4]1In the process, an antineutrino (ν) is also produced. The existence of the antineutrino was inferred
by the fact that the electron did not come out always with the same energy. There had to be anotherparticle responsible for this discrepancy but left no footprint, so it had to be neutral.[5]
2An antiparticle is a particle with the same mass but the opposite quantum numbers (electric charge,lepton number, etc.).[5]
Chapter 1. Introduction to Biomedical Imaging 4
• Gamma Decay. This decay produces electromagnetic radiation in the frequency of
gamma rays. The difference with other types of electromagnetic waves is the energy
and thus the frequency of the photons. It always comes alongside with another
type of decay. Because of their high energies, gamma rays penetrate large amounts
of matter before they are completely absorbed, which makes them interesting for
medical applications as diagnosis and treatment. When a positron encounters an
electron, they annihilate and produce a pair of photons of 0.511 MeV.
e+ + e− → γ + γ (1.4)
This process is used in Positron Emission Tomography (PET) for the diagnosis of
diseases by tracing the metabolical activity in certain parts of the body (Figure
1.2).
Figure 1.2: PET of a physiologically normal patient compared with that of a patientsuffering Alzheimer’s disease.[6]
These types of radiation are often referred to as ionizing radiation because all of them can
change the electrical charge of an atom, which may be the cause of damage to biological
tissues. This process is used to measure the presence of radiation with a Geiger counter.
Alpha particles are blocked by a sheet of paper, while beta and gamma radiation pass
through a paper barrier. Beta radiation is blocked with a thin piece of aluminum and
gamma rays can be attenuated with thick pieces of lead. Alpha decay has the shortest
range and can be easily stopped by skin.[7]
1.2 X-ray Production
X-ray radiation is the most widely used type of radiation for medical diagnosis. X-rays
are electromagnetic waves, with energies between 120 eV and 120 keV. In an X-ray tube,
electrons are thermally emitted from a filament (cathode) and are accelerated towards
Chapter 1. Introduction to Biomedical Imaging 5
the anode by a high voltage (of the order of 50 kV). When the electrons reach the target
in the anode, the Coulomb interaction with the nucleus of the atoms in the material
causes a continuum spectrum of electromagnetic radiation, called bremsstrahlung. The
maximum energy of the emitted X-rays is proportional to the voltage of the tube and
the spectrum covers energies up to this value, as an electron can interact with many
atoms in the target, until all its kinetic energy is converted into photons (the energy
that the target gains during the interactions can be neglected).
The difference in the kinetic energy of the electron before and after the interaction should
be equal to the energy of the emitted photon.
hν =hcλ
= K −K ′ (1.5)
The maximum energy that can be converted equals the energy gained by the electron
when it is accelerated by the voltage V (K − K ′ = eV ), from which the minimum
wavelength of the photons can be obtained.
λmin =hceV
(1.6)
� �
�
���
��
��
��������
����� �
���
Figure 1.3: (a) Schematic drawing of an X-ray tube in which the electrons are accel-erated by the voltage V, from the cathode C towards the anode A. (b) X-ray spectrum
showing bremsstrahlung and characteristic lines Kα and Kβ .[8]
The energetic electrons can excite the atoms of the anode target and when they return
to the ground state, a photon is emitted. The energy of this photon depends on the
difference of energies between the initial and final state of the electron and it falls in
Chapter 1. Introduction to Biomedical Imaging 6
the range of X-rays. The characterisitc lines of the material of the anode arise from the
fact that the intensity of X-rays is considerably larger than the bremsstrahlung at these
energies (see Figure 1.3b).[8]
1.3 Measurement of Radiation
Measurement of radiation can be a challenging task as we are not able to detect its
presence without specialized instruments. Different types of radiation usually come
together and are difficult to distinguish from each other. In that way, it is possible to
measure (1) the number of disintegrations per unit time in a sample, (2) the electric
effect this radiation has on atoms or ions, (3) the energy deposited to a sample and (4)
the biological effects that a dose has on a tissue.
The activity, or number of disintegrations that occur per unit time, regardless of the
source, is measured in Becquerel (Bq) or in Curie (Ci). One Becquerel (S.I. unit) equals
to one disintegration per second. However, this unit is not very convenient, so the Curie
is introduced. One Curie equals 3.7× 1010 Bq.
The exposure is related to the capability of ionizing certain material. The interaction
of radiation with atoms can cause a change in the net charge of the atoms and the
measurement of this charge can indicate how energetic the radiation is. The unit is
the Rontgen (R), which equals the amount of radiation needed to create a charge of
2.58× 10−4 C per kg of air.
The absorbed dose is a direct measure of the amount of energy delivered to a sample
by any kind of radiation. It can be measured in rad or Grays (Gy, S.I. unit). One rad
equals to 0.01 joule absorbed by one kg of material. One gray equals 100 rads, or 1 joule
absorbed per kg.[4]
The measurement of dose equivalent takes into account the effects of radiation in biolog-
ical tissues. This effect varies among the different types of radiation, even if they have
the same energy. This is expressed by the following equation:
DE = AD ×WR (1.7)
where AD is absorbed dose. If AD is in Gy, the unit of DE is the Sievert (Sv, S.I. unit).
If AD is in rad, the unit of DE is the Rontgen equivalent man (rem). It can be easily
seen that one Sv equals 100 rems. The weighting factor WR depends on the type of
radiation. For X-rays, gamma rays and beta particles it is equal to one and for alpha
particles to 10.
Chapter 1. Introduction to Biomedical Imaging 7
The biological effect of radiation depends also on the tissue that is affected. The effective
dose takes account of the affection to different organs:
ED = AD ×WT (1.8)
The effective dose (ED) is given in Gy, as the AD. The weighting factor WT is different
for each organ, defining the factor for the total body as equal to one. Table 1.1 lists the
weighting factors for different tissues.[9]
Tissue WT
Gonads 0.25Breast 0.15
Red Bone Marrow 0.12Lung 0.12
Thyroid 0.03Bone Surface 0.03Remainder* 0.03Total Body 1
Table 1.1: Weighting factors for different tissues. *For the remaining organs, a factorof 0.06 is used for each of the five organs receiving the highest dose.[10]
1.4 Detection of Radiation
The detection of radiation is a very important issue in the field of biomedical imaging
and is the one that presents the most difficulties. It must be clear that the detection of
radiation is based on its effect, as the darkening of a photographic film, the ionization of
a gas and the further measurement of produced charge, the production of luminiscence
in a crystal, or the production of charge carriers in a semiconductor. The first one, which
Rontgen used to give evidence of his discovery, is still used in some applications that
do not require the post-processing of images. The detection of radiation by ionization
is used in a Geiger counter, in which the ionization of a gas in a tube causes a current
to flow between the tube and a filament that goes trough the center of the tube. The
measurement of this electric current is an indirect measurement of radiation, as its
magnitude is proportional to the amount of incident radiation. The last mechanism is
the most widely used for medical imaging applications.
The next section presents a discussion of some instruments used to detect radiation with
imaging purposes. It is not intended to be a complete review of all of them, but a brief
description of some devices that have influenced the development of this field.
Chapter 1. Introduction to Biomedical Imaging 8
1.4.1 Scintillation Detectors
Scintillation detectors, which can be used with all types of radiation, use the production
of light in some crystals, such as zinc sulfide or sodium iodine, upon the deposition of
the energy carried by the incoming particles. The amount of light emitted in the crystal
is proportional to the energy of the particle that caused it. The scintillation crystal
can be placed next to an electronic device called a photomultiplier, that converts the
photons to electric pulses. The height of the pulses will be proportional to the energy
of the incident particle. These pulses can then be amplified, analized and stored, thus
giving information of the amount of radiation and its nature. The pulses have to be
amplified, as they are low voltage pulses at the end of the photomultiplier. This can
cause the background effects to affect the quality of the detection. The analysis of the
signal gives the possibility to distinguish radiation comming from different radionuclides,
an important issue in some medical applications. The user can program the electronic
device to show only the measurements that correspond to certain energies.[4]
In imaging devices, the signal of the photomultiplier is used to determine spatially the
origin of the radioactive source. In spite of the distortion that can affect the signal, due
to the fact that the light source lies in a three dimensional array and not bidimensional
as in other detectors, the introduction of scintillation devices resulted in an increase
of sensitivity and thus in the quality of the images. However, on recent years several
devices that give more spatial resolution than scintillators have been developed.
1.4.2 Charge-Coupled Devices
Charge-coupled devices (CCD) use a scintillator crystal or a semiconductor material
to convert the incident ray into a charge pulse that can be then measured in the shift
register, or the CCD. The detector is divided in small pixels that work indepedently to
collect and measure charge. In a one-dimensional row of pixels, each of them is divided
into three parts (i.e. φ1, φ2 and φ3) as shown in Figure 1.4. The applied voltage on each
division of one pixel is changed during the process of detection to move the accumulated
charge towards the end of the pixel line. During the first step, the voltage in φ1 and φ3
is kept high, while the voltage in φ2 is low. The accumulated charge in the high voltage
bins moves and is collected in the potential well defined by the voltage of φ2. After that,
the voltage of φ3 is lowered, and the charge in the potential well distributes in the bins
φ2 and φ3. Next, the voltage in φ2 is raised, causing all the charge to move to the φ3
bin. The voltage in φ1 is kept high to avoid the difussion of charge carriers from pixel
to pixel. This process is repeated and the charge that gets to the end of the pixel line
is measured and recorded.[11]
Chapter 1. Introduction to Biomedical Imaging 9
���������
�� �� ��� ���� �� ��
������
������
������
�
�
�
�
�
�
Figure 1.4: Schematic diagram of a CCD detector. Each pixel is divided in threeparts in which the absorbed charge moves to be measured.[11]
A two-dimensional array of pixels can be created by tiling several one-dimensional arrays,
working independently as described above. In this way, a two-dimensional image is
generated and can then be analyzed, processed and stored.
The advantage of CCD devices is the high resolution that can be achieved. The pixels
can have a size of the order of tens of microns, making this kind of detector suitable for
sensitive applications. The largest commercial CCD detector is 4K × 7K pixels, each
one measuring 12 µm × 12 µm (see Figure 1.5).
Figure 1.5: Largest commercial CCD with a sensitive area of 86 mm × 49 mm.[1]
1.4.3 Fast CT Scanners
The combination of high quality scintillator crystals and silicon photodiodes with a fast
gantry results in a fast Computed Tomography (CT) scanner, in which an image of the
whole body can be taken in about 25 seconds. The detector is able to take 64 slices per
Chapter 1. Introduction to Biomedical Imaging 10
revolution and the X-ray generator makes one complete rotation in 0.33 seconds. Besides
that, the X-ray tube has an advanced cooling system that uses oil in direct contact with
the anode, which allows to scan for prolonged periods (i.e. 25 s). The speed of this
system gives the possibility to reconstruct a beating heart, a difficult task due to the
fast movements of the imaged specimen.[1]
1.4.4 Energy-Resolving Methods
Radiography is based on the difference of absorption of X-rays in different tissues. The
amount of radiation that is absorbed by one material depends on its energy. For one
spectrum it is possible to have equal absorption with different materials, but this is no
longer true with another spectrum. The dependence of absorption with the incident
energy can be used to discriminate among different kinds of tissue. It is also possible to
enhance the contrast of an image by using this method.
An example of an Energy-Resolving Method (ERM) is shown in Figure 1.6. The addition
of images taken with X-rays of different energies, weighted by the appropriate factors,
makes it possible to obtain an image of the soft tissues. In a similar way, an image of the
bones can be obtained. The ERMs have the power of resolving very small malformations
or calcifications that are surrounded by different tissues, such as breast cancer.[1] This
task would be difficult to accomplish by any other method.
Figure 1.6: Radiographies taken with low (a) and high (b) energy spectra. An algo-rithm that uses both radiographies allows to visualize soft tissues (c) or bones (d) in
one image.[1]
1.4.5 Single-Photon Detection
In the common detectors it is required to integrate the signal and perform an analog-to-
digital conversion to handle information at a later point. However, quantum detectors
count single events, which results in a higher detective quantum efficiency (DQE), a
measurement of the quality of the resulting image. Single counting detectors work almost
Chapter 1. Introduction to Biomedical Imaging 11
without noise signal, because they are able to discriminate among different signals and
thus work over the noise level. Another advantage is the possibility of counting single
photons of different energies. There is one limitation that has to be considered with this
type of detectors. The amplifiers, discriminators and counters must be small enough to
fit in a detector containing a vast amount of pixels. The frequency at which the detector
counts is a limiting factor, as well.
Recent investigations of a chip that combines energy-resolving capabilities with single
photon counting, directed towards medical applications have given promising results.
The Medipix2 chip, developed at CERN, is a read-out matrix that, coupled to a semi-
conductor, allows the detection of charged particles and photons. It has a small effective
area of 14 mm × 14 mm, which, combined with its high resolution, makes it perfect
for the production of micro-radiographies, but it is also possible to tile several chips for
larger applications.[1]
Chapter 2
The Medipix2 Detector
The Medipix2 is based on the design of the Medipix1, the first approach to single photon
detection by the CERN Collaboration. The images taken with a Medipix1 showed
very good contrast and were free of noise, an advantage over other types of detectors.
However, it had some limitations that were overcome by the Medipix2. Among these
limitations was the size of the pixel (170 µm × 170 µm), that affected the spatial
resolution of the detector. The new detector had to have smaller pixels to increase the
granularity of the images. Another limitation was the lack of sensitivity to negative
impulses, thus the bias voltage had to be set to accumulate holes in the chip. The
compensation for leakage current was done in each column and for better results it had
to be done in each pixel. The dead area around the detector had to be decreased, to open
the possibility of tiling chips to have a detector with a larger sensitive area. To avoid
noise, the analog biasing had to be done in each pixel instead of being done externally.
The Medipix1 had only one discriminator, so it was not possible to set an energy window
for the detected particles. This had to be improved by adding another discriminator to
allow the setting of an energy window.
These improvements were implemented in the Medipix2, a 16120 µm × 14111 µm chip,
in an array of 256 × 256 pixels, each one of 55 µm × 55 µm, with a total sensitive area
of 1.982 cm2 (see Figure 2.1). The non sensitive area of the chip measures 2040 µm ×
14111 µm and has the wire-bonding pads, biasing Digital to Analog Converters (DACs)
and control logic.[13]
2.1 The Medipix2 Chip
When a charged particle or a photon incides on the semiconductor detector, an electron
hole pair is created. The bias voltage creates a high electric field in the material that
12
Chapter 2. The Medipix2 Detector 13
�������
����� ��
��
�����
� !"
�#
� !"
�$%&'()�%*+,�"*�$%-$%
'./()�%*012%*"3�4%*56��2%67
&&'8()�%*9�:6;*+�;$<
#(=
&&'8()�%*9�:6;*+�;$<
#(=
>?*8()�%*!@+2
&&'8()�%*9�:6;*+�;$<
#(>
&&'8()�%*9�:6;*+�;$<
#(>
&&'8()�%*9�:6;*+�;$<
#('..
&&'8()�%*9�:6;*+�;$<
#('..
����
���
� !"#
$%�&
''&(
)
Figure 2.1: Plan of the Medipix2 showing the sensitive area (blue), the non-sensitivearea (green) and its dimensions.[12]
separates the electron from the hole, drifting one of them towards the counting chip.
The charge is accumulated and amplified and it is translated into a voltage. This voltage
is compared with two thresholds and if it falls between the two thresholds, a counter in
incremented. This is done in every pixel of the matrix, giving rise to a quanta-counting
detector with an energy-resolving property.
2.1.1 The Analog Part
The analog part of the pixel (in red in Figure 2.2a) has a test capacitance, the Charge
Sensitive Amplifier (CSA) and the two discriminators. The input is the current pulse
that comes from the semiconductor detector upon the interaction of a photon or a
charged particle with the crystal.
The test capacitance is used to input a known amount of charge to the pixel to test the
settings. The CSA integrates the input charge and shapes it. It also has a compensation
for positive or negative leakage currents. The two discriminators are equal, but the
threshold voltage on each of them can be set individually. The high and low thresholds
are set by an 8-bit DAC for all the pixels of the matrix. A 3-bit DAC is used in each
discriminator branch of each pixel to minimize the threshold variation due to differences
Chapter 2. The Medipix2 Detector 14
�� ��
����
� � � !
"
�� ��
����
�� �� �� !!
""
���
!"#
$%%
&'()!*�+!,*
-./!"*.0
1234*
�*."*
$."*)!*
$."*51234*
67"8
'()!*5$%%5�9:
64;
64;
�<8=-.79
>0.?!@4"5>!;.<
A.;*5 >!;.<
�@2,B()!*"5>�-
>@<70 !*C
D
�+4**.0
'()!*5$%D5�9:
!"#
$%D
��� !" #$"$%�
���
� �
Figure 2.2: (a) Schematic of one Medipix2 pixel showing the analog (red) and dig-ital (blue) parts. (b) Pixel layout showing dimensions and the bump bond (green) inwhich the chip is connected to the semiconductor detector. (1) Preamplifier; (2) HighThreshold Discriminator; (3) Low Threshold Discriminator; (4) 8-bit Pixel Configura-
tion Register; (5) DDL; (6) SR/C (13-bit shift register and logic).[12]
Chapter 2. The Medipix2 Detector 15
in the transistors and power consumption. This DAC controls the amount of current
that is added to the output of the discriminator and is unique for each threshold and
each pixel. At the end of each discriminator there is a mask bit that allows to discard
the information from noisy or defective pixels.[13]
2.1.2 The Digital Part
The digital part of the pixel (in blue in Figure 2.2a) has the Double Discriminator Logic
(DDL), the Shift Register Counter (SR/C) and the Pixel Configuration Register (PCR).
The DDL uses the output of each discriminator branch and its output is used as a clock
to increment the counter of the SR/C. A global clock is used to shift the data from
pixel to pixel, to set the 8 configuration bits or to read the 13 pixel counts bits with the
Shutter and Conf signal controls.
The DDL compares the output of each discriminator and depending on the configuration
of the thresholds it uses a clock to count bits. The detector can be set in a single threshold
mode, by setting the high threshold (THH) lower than the low threshold (THL). In this
setting, the pulses with a voltage higher than THL are counted. On the window mode,
an energy window is set with the THL and THH voltages. The pulses with a voltage
higher than THL and lower than THH are counted. The rest of the pulses are discarded.
If the Shutter and Conf are low, the SR/C counts each clock impulse comming from
the DDL, up to 8001 counts. If the Shutter is high, a global clock shifts the data from
pixel to pixel. In this mode, if Conf is low, the information of the counts of each pixel
is read out. On the contrary, if the Conf is high, the PCR sets the configuration of the
pixel.[12]
The PRC is an 8-bit memory that stores the configuration of the pixel. The configuration
it stores is the THL (3 bits), THH (3 bits), one test bit and one mask bit.
Figure 2.2b shows the pixel layout. The bump bonding aperture, shaded in green, has
a diameter of 20 µm. It lies upon the analog part of the pixel and communicates the
semiconductor detector with the chip. The detector is assembled to the chip using an
eutectic tin-lead solder process.[13]
2.1.3 The Chip Periphery
The periphery is the non-sensitive area of the chip, which provides the analog biasing
and the digital controls of the matrix. It has the width of the matrix and the height
provides enough clearance for the wire bonding process.
Chapter 2. The Medipix2 Detector 16
The analog part of the periphery has 13 8-bit Digital to Analog Converters (DACs),
which provide bias voltages and currents to each of the pixels of the matrix. The DACs
are set with an 8-bit code that can be set in different ranges of values for each of the
controlled parameters. The principal DACs are the FBK (set to 128), the Ikrum (set to
20) and the THL and THH. The FBK and the Ikrum are parameters of the CSA that
control the amplification (FBK voltage) and leakage current effects (Ikrum current).
The effect of rising the FBK is to lower the voltage equivalent to a charge impulse,
affecting also the noise level. This DAC is important when working with paticles that
deposit a large amount of energy that saturates the pulse. The Ikrum controls the rise
time of the pulse. A higher Ikrum gives a shorter time, thus a higher voltage pulse.[14]
The test pulse is also controlled by the analog periphery of the chip. It sets the amount
of charge that will be injected to each pixel, the number of pulses and the rise and fall
time of the input voltage.
The digital part of the periphery does the reading and writing of the pixel matrix and
the loading of the DAC digital codes. The Low Voltage Differential Signaling (LVDS)
is the pathway to data transmission to and from the chip.[13]
2.2 The Mpix2MXR20
The Mpix2MXR20 is the redesign of the Medipix2 that adresses the correction of some
limitations of the Medipix2. The Medipix2 solved the problems that the Medipix1 had by
using 0.25 µm CMOS technology. It had very small power consumption, fast operation
and tiling of several chips was possible. However, there were still some problems that
limited the practical applications of the chip.
The most important limitations of the Medipix2 were the following:
• The voltage DACs had a high temperature dependence and a lack of linearity.
• There were variations in the analog buffers from pixel to pixel, limiting the func-
tionality of the test pulse.
• The counter was reinitialized when it reached 8001 counts and there were no means
to know if this happened, which generated a loss of information.
• The radiation hardness (i.e. 10 krad) was not as high as expected with the tech-
nology used. This means that the integrated dose that caused a clear variation in
the THL equalization mask was an order of magnitude lower than expected.
Chapter 2. The Medipix2 Detector 17
���
����
�
�� �������
��������
�����
�����
���� ��
����������
����
�� ��������
��
��
�!�"����
#��$�%���#� �!
&� �� #� �!
�%��'� ����#��
#%!�� ��(
������
��)
�������
�� ����)����
����
�)
������
*$�� �!%+�%���%!
Figure 2.3: Schematic of one Mpix2MXR20 pixel showing the analog (red) and digital(blue) parts.[13]
Besides solving the previous limitations, the redesign of the Medipix2 modified some
parts of the pixel and periphery electronics. The dimensions of the chip and pixel cell
and the readout architectrure were not modified to use the same sensor materials and
readout systems designed for the Medipix2. At a pixel level, NMOS enclosed layout
transistors were used in the sensitive nodes to eliminate the effect of parasitic currents
due to radiation. A local mirror was added to generate the Ikrum/2 used for leakage
current compensation. In this way, the dispersion of the threshold was improved. The
discriminator block was simplified and the mask bit was changed to the digital side
(see Figure 2.3). In the digital side, an overflow bit was added and the SR/C was
slightly changed. The counter was changed to have a dynamic range of 11810 counts
in acquisition mode (Shutter low). If this number is reached during an acquisition, the
overflow bit is modified and the logic changes the pixel from acquisition to read-out
mode (Shutter high).
The periphery was modified to improve the DACs temperature dependence, linearity, use
of test pulses and readout speed. The dimensions of the periphery were not changed.
The most important modification in the chip periphery was the improvement of the
DACs stability and temperature dependence by the implementation of a circuit that
delivers a temperature independent current. The periphery of the Mpix2MXR20 has 11
8-bit current and voltage DACs to control the analog and digital parts of each pixel. It
also has two 14-bit voltage DACs to set the thresholds in a more precise way. Each of
them has a linear 10-bit DAC to set the fine threshold and a 4-bit DAC to set the coarse
threshold.[13] With this two DACs, the low and high threshold can be set in a wide
range of values with the THL coarse and THH coarse and then tuned more precisely
with the THL and THH. The latter present high linearity with the voltage, which is
useful to detect particles or photons with a good energy resolution.
Chapter 2. The Medipix2 Detector 18
The Mpix2MXR20 is the chip that was used for all the measurements. From now on
the Mpix2MXR20 will be referred to as Medipix2.
2.3 The USB Interface
The USB Interface, developed at the Institute of Experimental and Applied Physics
(IEAP) of the Czech Technical University in Prague is needed to connect the Medipix2
to a PC. It is an alternative to the MUROS2 interface, developed at the National Insti-
tute for Nuclear Physics and High Energy Physics (NIKHEF) in the Netherlands. The
USB Interface allows to readout the pixel matrix and to control the acquisition and its
parameters.
The USB Interface has an internal source of variable bias voltage with leakage current
monitor that allows to set voltages from 5 to 100 V. This is the voltage along the
semiconductor sensor that separates the electron hole pair for the detection of one of
them. Besdies the independence of an external power source, the USB Interface is more
portable than the MUROS2, an advantage for some applications. The firmware can be
flashed directly via USB and it has a PCB module for additional applications.[15]
Figure 2.4: The USB Interface (a) in the box and (b) out of the box.[15]
2.4 The Pixelman Software
The Pixelman software, developed at the IEAP controls the acquisition of Medipix2 via
the USB Interface or the MUROS2. It is an alternative to the Medisoft4 sotware, de-
veloped at the University of Napoli. The plugins allow to perform operations additional
to the acquisition, such as coincidence control, beam hardening correction and more
complex measurements.
Chapter 2. The Medipix2 Detector 19
Figure 2.5: The Pixelman principal window in which the parameters of the acquisitioncan be controlled.[15]
The principal window has the acquisition options. The acquisition type, time of ac-
quisition, number of acquisitions, and Spacing can be set before the acquisition begins.
The Spacing is an important parameter of the acquisition. It controls the number of
pixels that are active during the acquisiton to avoid coupling effects among neighboring
pixels. The number of Spacing corresponds to the number of inactive pixels between
the ones that are counting. The type of output and the file extension can be changed in
this window. This window also provides access to the Preview window, Settings, DACs
control panel and DACs Scan.
Figure 2.6: The Pixelman Preview window in which the acquired data is displayed.[15]
Chapter 2. The Medipix2 Detector 20
The Preview window displays the data that is read from the Medipix2 in a matrix
with different color schemes. It also presents a histogram that gives the information of
number of pixels with a certain count and statistical data of the displayed values. The
equalization masks, test bits and mask bits can be visualized in the Preview window.
The software allows to save the data in image or text files.
The user can control the settings of the Medipix2 in the Settings window. The polarity
of the device can be selected to collect holes (Positive) or electrons (Negative) and the
Bias Voltage can be set up to 100 V. A digital test can be performed to know the number
of good and bad pixels. The program writes and reads a random matrix and compares
the data to know how many pixels are working. The DACs can be controlled in this
window, as well as in the DACs Control Panel, which can be accessed from the principal
window.
The Tools menu in the principal window gives access to the DACs Scan, the DACs
Control Panel, the Threshold Equalization and the Beam Hardening Module.
Chapter 3
Preliminary Measurements and
Energy Calibration
This chapter describes the measurements done with the Medipix2 detector prior to the
acquisition of radiographies. The first procedure was the threshold equalization, in which
the two thresholds of each pixel are corrected with a 3-bit DAC. This allows to have
a more unified threshold throughout the matrix. After that, a DAC scan of the THL
was done. From this scan some propierties of the detector can be obtained, such as the
effective threshold and the pixel linearity.
The next part presents the energy calibration of the detector using the procedure de-
scribed in [14, 16]. The objective of this calibration is to find the correspondence between
DAC threshold value and actual energy. This is an important procedure to take advan-
tage of the energy window feature of the Medipix2.
3.1 Threshold Equalization
In the analog part of each pixel there are two discriminators in which the incoming
pulse is compared with two voltages that constitute the energy window. Each branch
has a 3-bit DAC used to control the amount of current that is injected to the output of
the discriminator in order to minimize the threshold variation due to differences in the
transistors and power consumption. This 3-bit DAC is different for each branch and for
each pixel.
The threshold equalization is the procedure by which the 3-bit DAC value is set for the
THL and THH branch for each pixel. The first is the low threshold equalization, that
uses the noise floor to determine the vaule of THL DAC at which a pixel stops counting.
21
Chapter 3. Preliminary Measurements and Energy Calibration 22
This is done once with the 3-bit current DAC set to 000 (red distribution in Figure 3.1)
and then set to 111 (blue distribution in Figure 3.1). As the behavior of the threshold is
linear, an interpolation can be done to determine the 3-bit current DAC that minimizes
the dispersion (black distribution in Figure 3.1).
��� ��� ��� ��� ��� ��� ����
����
����
����
����
�����
�������� !"#$%"!"&� !"#$%
�&
!#$
Figure 3.1: Distribution for the low threshold equalization with 000 correction (red),111 correction (blue) and after the equalization (black).
In this case, the standard deviation was 10.38 and 9.50 (in DAC units) for 000 and 111
correction, respectively. After the THL equalization, the standard deviation was 2.43.
The result of this procedure is a 256×256 matrix with values between 0 and 7, that
represent the 3-bit current DAC that corrects each pixel’s low threshold.
After the THL is equalized, the high threshold equalization has to be done if an energy
window is to be set. This was done following the procedure described in [17]. The
Double Discriminator Logic of the Medipix2 is such that if the THH is below the THL,
the detector works in a single threshold mode and all the pulses with voltage larger than
the THL are counted. In the double threshold mode the THL is below the THH and
the pulses that lie in the energy window are counted. The equalization of the THH uses
this transition from single to dual energy mode, setting the THL above the noise level
and pulses with voltage higher than the THL. The THH was scanned in the vicinity of
the THL such that when it was below the THL, the pulses were counted and when it
was above the THL, and worked in double threshold mode, the pixel stopped counting
the pulses.
Chapter 3. Preliminary Measurements and Energy Calibration 23
��� ��� ��� ��� ��� ��� ��� ��� ��� ����
����
����
����
����
����
����
���
���������� !"�#�$%!"&'
�($
%"&
Figure 3.2: Distribution for the high threshold equalization with 000 correction (red),111 correction (blue) and after the equalization (black).
This THH scan was done with the 3-bit current DAC set to 000 (red distribution in Fig-
ure 3.2) and then set to 111 (blue distribution in Figure 3.2). The same interpolation
technique was used to get the narrower distribution of the corrected pixels (black distri-
bution in Figure 3.2). The standard deviation was 15.14 and 12.88 (in DAC units) for
000 and 111 correction, respectively. After the THH equalization, the standard deviation
was 3.82.
The equalization procedure has the possibility of masking the pixels that are far from
the average. The user can set the number of standard deviations from which the pixels
begin to be masked. In this equalization, the pixels lying three standard deviations
from the mean were masked. After the THL and THH equalizations, the total number
of masked pixels was 647.
3.2 DACs Scan
The DACs Scan is a useful plugin to perform a scan of any DAC value. The electrical
characterization of the pixel was done with this utility, following the procedure described
in [13]. The test pulse was used to inject pulses to eight pixels along the matrix (while
the rest was masked). The THL was scanned over values ranging from above the pulse
voltage to below the noise level. The THH was set at half the distance between the
pulse height and the noise level. The graph obtained is shown in Figure 3.3.
Chapter 3. Preliminary Measurements and Energy Calibration 24
��� ��� ��� ��� ��� ��� ��� ��� ���
�
����
����
����
����
�����
�����
�������� !"#$%"!"&�'($%)*
+,!
(�-
$.,/��
0'(%
Figure 3.3: THL DAC scan with 4000 test pulses of 0.1 V.
The first information that can be obtained from the graph is that a larger THL DAC
corresponds to a smaller voltage value. This is the reason for the noise level to be at
a larger DAC value than the pulse level. The first rise occurs when the THL crosses
the test pulse height, going towards the noise level. The number of counts in the active
pixels rises up to the number of pulses injected in each pixel. In this region the Medipix2
works in single threshold mode. Thus, the Counts value remains constant until the THL
crosses the THH. In this moment the detector passes to a double threshold mode, giving
no counts, as the pulse height is above the energy window. The next rise comes when
the THL crosses the noise level.
This graph also gives information of the effective threshold, linearity, gain and noise.
The effective threshold can be extracted using the s-curve method, that states that the
effective threshold is at 50% of the rising or falling edge of the curve.[13] The gain is
the distance between the effective threshold and the noise floor. If this scan is done
for different voltages of the test pulse, an effective threshold can be found for each one.
The linearity of the threshold with the voltage can be measured with these values. The
measurement of gain and noise can be done for different values of the Preamp and Ikrum
DACs. A Preamp of 128 and Ikrum of 20 is found to give an optimal combination of
gain and noise.[13]
All these measurements were done with an active matrix of eight pixels to avoid coupling
among pixels. The Pixelman software has a utility that controls the number of active
pixels during an acquisition. The Spacing is the number of masked pixels between active
Chapter 3. Preliminary Measurements and Energy Calibration 25
pixels during a subacquisition. This is used to minimize the coupling effects from pixel
to pixel. The number of subacquisitions is equal to 2S where S is the Spacing. A
Spacing of 4 has a good balance between pixel coupling and time required to complete
the acquisition.[15]
3.3 Energy Calibration
The next step was the energy calibration, for which the procedure described in [14, 17]
was used. The Medipix2 is illuminated with a known X-ray or gamma ray spectrum and
a THL DAC scan is performed. The points at which the rate of change of the Counts
has peaks are the points at which the characteristic lines of the source lie.
Before calibrating the Medipix2, the spectrum of each tube was measured with the
PHYWE X-ray unit (see Appendix C). Bragg’s Law states that maximum intensities of
light inciding at an angle θ to a crystal of lattice constant d occur at integral number n
of wavelengths λ,[18] such that:
2d sin θ = nλ (3.1)
An expression that relates the energy of the reflected light with the angle of reflection
can be found from Bragg’s Law:
E =nhc
2d sin θ(3.2)
The PHYWE X-ray unit was used to measure the intensity of X-rays at angles ranging
from 3◦ to 30◦, covering the first order of diffraction (i.e. n = 1), for both tubes with a
voltage of 35 kV and current of 1 mA. Figure 3.4 has the results of the measurement of
the spectrum of the Cu and Mo X-ray tubes. The characteristic lines, found with (3.2),
are:
Element/Line λ [pm] Energy [keV]Cu Kα 154.794 8.009Cu Kβ 139.746 8.872Mo Kα 71.330 17.382Mo Kβ 63.012 19.676
Table 3.1: Characteristic lines for Cu and Mo X-ray tubes.
The Medipix2 was illuminated with each tube and a THL DAC scan was performed. The
THL was scanned towards the noise level with a step of 3 DAC values. The scan with
the Cu tube shows only one peak (see Figure 3.5a), corresponding to an energy between
Chapter 3. Preliminary Measurements and Energy Calibration 26
�� �� �� �� ��� ��� ��� ��� ��� ��� ��� ���
�
����
����
����
����
����
����
���
����
���
�����
������
��
!" #$!%��&'( !#)#"&�
�� �� �� �� ��� ��� ���
�
���
���
���
���
����
����
����
����
����
������
!"
#$%&
$'(%)��*
+,$%
'-'&
*�
Figure 3.4: Spectra for Cu (a) and Mo (b) X-ray tubes measured using Bragg’sDiffraction.
Chapter 3. Preliminary Measurements and Energy Calibration 27
��� ��� ��� ��� ��� ��� ����
��
��
��
��
��
�
�
��
��������� !"#�$%!"&'
�'
(�)*$+',(����'
��� ��� ��� ��� ��� ��� ���
�
�
�
�
��
��
��
��
��
��
��
��
��
��������� !"#$! !%�&'#$()
�")
*�+,&-).*����)
Figure 3.5: Derivative of the flux found with the THL DAC scan with Cu (a) andMo (b) X-ray tube.
Chapter 3. Preliminary Measurements and Energy Calibration 28
the Kα and Kβ lines. The scan with the Mo tube shows two peaks that correspond to
the two characteristic lines (see Figure 3.5b). This gives rise to three values of energy
with their corresponding THL value.
THL DAC Energy [keV]317 8.44256 17.382250 19.676
Table 3.2: Values obtained from energy calibration with Cu and Mo X-ray tubes.
3.4 Validation of Calibration
� �
�
�
�
� !"!#$%#&'()$*+&,-
� !"
# $%
&�
Figure 3.6: Cross sections for Ni (blue), Cu (black) and Zn (red) showing the K lineat 8.333, 8.979 and 9.659 keV, respectively.[19]
The calibration obtained with the characteristic lines of the Cu and Mo X-ray tubes
was compared with THL DAC values obtained from the K lines of three elements.[19]
Nickel, copper and zinc foils were used as absorbers in the Mo X-ray tube aperture, with
a voltage of 17 kV and current of 1 mA, using the bremsstrahlung of the tube. The
cross section of these materials shows the K line at energies below 10 keV. The rise in
the cross section at these energies should be visible if the element is illuminated with
X-rays, as suggested by Figure 3.6.
Figure 3.7 shows the derivative of the flux obtained from the THL DAC scan. The peaks
of each material can be related with the respective K line energy (see Table 3.3).
Chapter 3. Preliminary Measurements and Energy Calibration 29
��� ��� ��� ��� ��� ��� ��� ��� ��� ��� ��� ��� ��� ��� ��� ��� ���
��
��
��
�
��
��
��
��
��
��
��
��
��
���������� !"�#$ !%&
'�()#
*&+'����&
Figure 3.7: Derivative of the flux found from the THL DAC scan with Ni (blue), Cu(black) and Zn (red) foils, illuminated with the bremsstrahlung from a Mo X-ray tube
at 17 kV.
THL DAC Energy [keV]318 8.333314 8.979310 9.659
Table 3.3: Values obtained from THL DAC scan with Ni, Cu and Zn foils.
A linear fit was done with the values of THL DAC and corresponding energies, as shown
in Figure 3.8. The linearity of the energy of the inciding photons with the Medipix2
threshold is evident from the fit.
E = 58.81574± 5.002− 0.15911± 0.01812× THL
R2 = 0.98720 (3.3)
Chapter 3. Preliminary Measurements and Energy Calibration 30
��� ��� ��� ��� ��� ��� �� � �� ��
�
�
�
�
�
�
��
��
��������� !"#�$%!"&'
(%)*#�+,)-
.
Figure 3.8: Energy calibration with Cu and Mo X-ray tubes (�) and validation withNi, Cu and Zn foils illuminated with X-rays (N).
Chapter 4
Radiographical Imaging with
Medipix2
When passing through a sample, the intensity of an X-ray beam is attenuated. Different
materials attenuate an X-ray beam in different amounts. These differences in attenuation
can be used to evaluate the internal structure of a sample, without having to look directly
at it.
This chapter describes the procedures done to obtain radiographies with Medipix2 semi-
conductor pixel detector. The first section decribes the experimental setup used to ob-
tain radiographical images with the PHYWE X-ray unit and the Medipix2. After that,
the flat field correction is presented with an example showing its possible contribution
to the quality of the radiographies. Then, the energy window feature of the Medipix2
is exploited to take radiographies at different energies, from which the need of beam
hardening correction arises. The beam hardening correction is then presented and some
examples that display a noticeable difference with raw and flat field-corrected images
are shown.
4.1 Experimental Setup
The Medipix2 was placed inside the PHYWE X-ray unit using the Medipix2 support
designed and constructed at UniAndes (see Appendix D for details). Figure 4.1 shows the
Medipix2 on the support, inside the PHYWE X-ray unit. For the following radiographies
the Mo tube with a voltage of 35 kV and current of 1 mA was used. The detector was
placed in front of the 5 mm collimator at a distance of 390 mm. An exposure time of
31
Chapter 4. Radiographical Imaging with Medipix2 32
Figure 4.1: Experimental setup used for radiographical imaging with Medipix2. Thefigure shows the PHYWE X-ray unit with a Mo tube and the Medipix2 on the support
designed and constructed at UniAndes.
2.5 seconds, divided in 5 acquisitions of 0.5 seconds each, was used to avoid the pixel
counter overflow and to have images with good statistics.
The sample was placed next to the detector. For this section, a sample support was not
used due to the dimensions of the sample. The active area of the Medipix2 of 14 mm
× 14 mm was not suitable to take a complete image of the sample if it was not placed
next to the detector’s window.
4.2 Flat Field Correction
The flat field correction is used to eliminate the distorition of the image due to the differ-
ence in efficiency among pixels. This correction can also compensate the negative effects
of a non-uniform radiation field. After correcting with flat field, a radiography displays
details of the sample with higher contrast and the statistical variation throughout the
image is reduced. The regions with little or no attenuation show the best results after
the flat field correction with an open beam image.
To correct an image with this method,[20] the raw image has to be multiplied by a
matrix C containing the correction factor for each pixel. These factors can be found
by illuminating the detector with the same parameters of the acquisition, but without
the sample in between. The correction factor for the i,j th pixel is given by cij = f/fij ,
where f is the mean value of all pixels and fij is the value of counts in the i,j th pixel
Chapter 4. Radiographical Imaging with Medipix2 33
of the open beam image. The corrected image, represented by a matrix V is given by
vij = cijrij , where rij is the value of counts in the i,j th pixel of the raw image R.
4.2.1 Foils Phantom
The foils phantom was constructed with aluminum foils with a thickness of 1.58 mm
each. The phantom had a part without foils (lowermost), one foil, two foils and three
foils (uppermost). The radiography was taken with an energy window from 8 to 35 keV.
Figure 4.2a shows the raw image and the respective histogram of pixel counts of the foils
phantom. Figure 4.2b shows the image and the respective histogram of pixel counts of
the foils phantom after the flat field correction with open beam.
��� ���
Figure 4.2: Comparison of raw (a) and flat field-corrected (b) image of foils phantomwith their respective histograms of pixel counts. The black points in both images are
masked pixels.
The first difference between the raw and corrected image is the visibility of the non-
uniform radiation field in the first one. On the other hand, the corrected image shows
little or no effect due to the non-uniform X-ray beam. The uniformity of the colors is
better in the corrected image, especially in the parts with less absorption (two lower
Chapter 4. Radiographical Imaging with Medipix2 34
regions). This is due to the fact that the correction was done with an open beam image.
If it were done with a flat field image of two or three aluminum foils, the two upper
regions would display better uniformity than the lower ones.[20]
The histograms of pixel counts of both images show three peaks that correspond to the
first three regions (from top to bottom) of the phantom. It is evident that the histogram
of the first image has a wider distribution for the three peaks than the second. The flat
field correction causes the distributions to be narrower, another sign of better uniformity
after the correction.
4.2.2 Capacitor
Figure 4.3: Photograph of capacitor.
The next radiographies were taken to the capacitor shown in Figure 4.3. The raw image
of the capacitor (Figure 4.4a) has a visible non-uniformity, which fades away after the
flat field correction with an open beam image. The color variation in the internal part
of the capacitor is better in the corrected (Figure 4.4b) than in the raw image. The
effect of the correction on the histogram is more evident in this case than in the foils
phantom. The two distributions appear more separated and with less deviation in the
second image, which translates into a more uniform image.
It is therefore clear that the flat field correction has to be taken into account to get the
best resolution and quality from the pixel detector. However, as will be shown in the next
sections, it is only part of the whole correction needed to obtain actual radiographies.
Chapter 4. Radiographical Imaging with Medipix2 35
��� ���
Figure 4.4: Comparison of raw (a) and flat field-corrected (b) image of capacitor withtheir respective histograms of pixel counts. The black points in both images are masked
pixels.
4.3 Radiographies at Different Energies
The absorption of radiation depends on the material through which it passes and also on
the energy of the inciding photons. Radiation of low energy is stopped in a higher amount
than that of higher energy. Therefore, if the detector has the possibility of counting only
the radiation with certain energy, it is possible to visualize different information with
different parts of the spectrum of a polychromatic X-ray tube (see Figure 1.6).
The energy window feature of the Medipix2 was used to take radiographies of the same
capacitor of last section (see Figure 4.3). The Mo X-ray tube was used with a voltage
of 35 kV and current of 1 mA. Each radiography is corrected with an open beam flat
field taken with the same parameters as the raw image.
The radiographies taken with 5 to 10 keV and 10 to 15 keV X-rays (Figures 4.5a and
4.5b) show very little contrast inside the capacitor. This is due to the high absorption of
the photons of low energies. It is not possible to distinguish among the different materials
Chapter 4. Radiographical Imaging with Medipix2 36
��� ��� ���
��� ��� ���
Figure 4.5: Radiographies with flat field correction of the capacitor of Figure 4.3 at(a) 5 to 10 keV, (b) 10 to 15 keV, (c) 15 to 20 keV, (d) 20 to 25 keV, (e) 25 to 30 keV
and (f) 30 to 35 keV.
with these spectra. Figures 4.5c and 4.5d, taken with 15 to 20 keV and 20 to 25 keV
X-rays, respectively, display better contrast among materials of different densities. In
Figures 4.5e and 4.5f, taken at 25 to 30 keV and 30 to 35 keV, respctively, the statistical
nature of radiation is evident. This effect is due to the low quantity of photons at these
energies. Even though the images are corrected with an open beam flat field, the last
two display high non-uniformity.
Taking radiographies at different energies gave complementary information of the studied
sample. Spectra of low energies can show differences among materials of low density.
The high quantity of low-energy photons gives images with very good statistics, which
drops at high energies. Higher energies are able to show differences among more dense
materials, but the low quantity of photons affects que quality of the radiographies. This
has to be compensated with higher exposure times. Despite the advantages that bring
an energy-resolving detector and the flat field correction, the radiographies obtained so
far present variations and are not suitable for tomographical purposes.
Chapter 4. Radiographical Imaging with Medipix2 37
4.4 Beam Hardening Correction
The attenuation of X-rays depends on the beam energy, as shown in the last section. In a
pixel detector used for transmission radiography the spectrum in each pixel is unique and
depends on the composition of the sample it passed through. The energy dependance
of the absorption causes the spectrum to be harder in the thicker parts. As each pixel
has a unique efficiency, a correction method has to be implemented in order to address
for the beam hardening issue.
The Signal to Thickness Calibration (STC) is based on the assumption that the sample is
composed of one material with different denisty from point to point.[20] The calibration
is done with flat absorbers of various thicknesses. An open beam image is taken with each
absorber, until the thickness of the sample is covered. The absorbers can be aluminum
or plastic foils. A set of values of count rates yk and thicknesses xk for each pixel is
used to find the equivalent thickness of a certain count rate by interpolation. If the raw
image has counts rij , the corrected image will have counts bij = h(rij), where h is the
calibration function. It is assumed that each point can be modeled by an exponential
function with offset.
y = h−1(x) = Akeakx + Ok (4.1)
The parameters of the exponential fit, Ak, ak and Ok are found for each pixel using the
set of measurements with absorbers. The value y is corrected by interpolation between
the two nearest calibration points, such that yk > y > yk+1. This value is given by:
h(y) =y − yk+1
ak(yk − yk+1)ln
y −Ok
Ak+
yk − yak+1(yk − yk+1)
lny −Ok+1
Ak+1(4.2)
The Pixelman software has a Beam Hardening plugin that computes the exponential fit
for every pixel from a stack of flat field images, taken with a set of absorbers of different
thickness. This configuration can be saved to perform the correction of the raw images
at a later time. The parameters of the acquisition and the experimental setup have to
be the same for the calibration and the acquisition of the raw images. The plugin has
an algorithm for correcting defective pixels by nearest-neighbor weighted average.[15]
Aluminum foils of 0.02 mm to 0.3 mm in 0.02 mm increments and 1.58 mm to 14.22
mm in 1.58 mm increments were used for the following corrections. Each foil or stack of
foils was placed in front of the X-ray tube aperture with the same parameters as before.
Chapter 4. Radiographical Imaging with Medipix2 38
4.4.1 Foils Phantom
The first difference between the raw image of the foils phantom (Figure 4.6a) and the
STC-corrected image (Figure 4.6b) is the change in uniformity in the lowermost part
(region with no foils and one foil). The other two regions do not show the same level of
correction, but this is due to the low amount of counts in these parts. The histogram
gives evidence of this correction by narrowing the deviation of the peaks. Moreover, the
distance between peaks becomes constant after the correction. In the first histogram
only three out of four peaks are visible (the small peak at zero counts corresponds to
masked pixels). In the histogram of the corrected image the four peaks, corresponding
to the four regions, are visible.
��� ���
Figure 4.6: Comparison of raw (a) and STC-corrected (b) image of foils phantomwith their respective histograms of pixel counts.
It is also noticeable that the corrected image presents no noisy or masked pixels. The
algorithm included in the beam hardening correction compensates very well this problem
by detecting pixels with zero counts (masked) and pixels with bad fitting curves (noisy).
Chapter 4. Radiographical Imaging with Medipix2 39
4.4.2 Capacitor
In the case of the capacitor, the STC-corrected image (Figure 4.7b) is clearly more
uniform than the raw image (Figure 4.7a). The difference beween the wall of the cylinder
and the internal material is clearly defined in the second image as well as the distinction
among the materials in the interface between the two wires and the capacitor. It is
possible to distinguish the wires from a round plastic part and the metal plates in the
interior of the capacitor.
The histogram of the raw image has a high frequency at the first and last bins. The
correction, besides inverting the histogram, decreases the separation between the two
peaks. The peak at the high pixel counts in the first histogram corresponds to the
narrow distribution around zero in the second histogram. This is a clear sign of better
uniformity, which is also visible in the open part of the corrected radiography.
��� ���
Figure 4.7: Comparison of raw (a) and STC-corrected (b) image of capacitor withtheir respective histograms of pixel counts.
In this chapter it was shown that, despite the raw radiographies give information of
the internal structure of an object, as long as the radiation is absorbed in a convenient
Chapter 4. Radiographical Imaging with Medipix2 40
way (not totally absorbed or totally unaffected), the images have to be corrected to
have an optimal quality. The flat field correction eliminates the differences in efficiency
of the pixels and is enough for monochromatic radiaton. However, the availability of
a monochromatic radiation source in practical applications of radiography is unusual.
The beam hardening correction has proved to eliminate the negative effects of a poly-
chromatic source of radiation. The energy window feature of the Medipix2 was used to
show the difference in absorption of different parts of the spectrum by the same sample,
although this is not the only application of this capability (see Chapter 1).
Chapter 5
Methods for Tomographical
Reconstruction
This chapter presents two approaches to the problem of reconstructing the cross section
of an object from its projections. The first approach is the mathematical formulation of
the problem by Allan Mcleod Cormack.[2, 3] It is based on the Fourier Slice Theorem
and is implemented with the use of the Fast Fourier Transform. The second approach
is an iterative algorithm that has proved to be useful when the number of projections is
low. Finally, a short comparison between the two methods is presented.
5.1 The Filtered Back Projection Method
The Filtered Back Projection (FBP) method is an analytic algorithm to solve the prob-
lem of reconstructing the slice of an object from its projections. The combination of line
integrals gives rise to the projection of an object. This definition is used to derive the
Fourier Slice Theorem which leads to the possiblility of reconstructing the cross section
of the object with the FBP algorithm, as presented in [21].
5.1.1 The Projection of an Object
The projection of an object arises from the evaluation of a parameter of the object along
a line. In this case, the parameter is the attenuation of X-rays as they pass though a
material.
41
Chapter 5. Methods for Tomographical Reconstruction 42
�
�
������
������
Figure 5.1: Geometry definition for tomographical reconstruction. The functionf(x, y) represents the object and P (θ, t) its projection.
Figure 5.1 shows an object with attenuation coefficient given by the function f(x, y) and
(θ, t) the two parameters that define the line integral given by:
P (θ, t) =∫
(θ,t)f(x, y)ds (5.1)
The function P (θ, t) is the Radon transform of the function f(x, y). The Radon trans-
form is related to the coefficient of transmission by:
P (θ, t) = − ln(
IIo
)
(5.2)
The projection is the combination of several line integrals, from which the object can be
reconstructed.
5.1.2 The Fourier Slice Theorem
The Fourier transform of the object function is given by:
F (u, v) =∫ ∞
−∞
∫ ∞
−∞
f(x, y)e−2πi(ux+vy)dxdy (5.3)
Chapter 5. Methods for Tomographical Reconstruction 43
The Fourier transform of the projection at an angle θ is given by:
S(θ, w) =∫ ∞
−∞
P (θ, t)e−2πiwtdt (5.4)
If the case θ = 0 is considered, setting v = 0 in (5.3) gives:
F (u, 0) =∫ ∞
−∞
[∫ ∞
−∞
f(x, y)dy]
e−2πiuxdx (5.5)
F (u, 0) =∫ ∞
−∞
P (θ = 0, x)e−2πiuxdx (5.6)
The integrals can be splitted and the term in square brackets is equal to the projection
at an angle θ = 0. The result is that the Fourier transform of the object function equals
the Fourier transform of the projection:
F (u, 0) = S(θ = 0, u) (5.7)
This particular result can be generalized to all (θ, t) leading to the Fourier Slice Theorem[21],
that reads:
The Fourier transform of a parallel projection of an image f(x, y) taken at
an angle θ gives a slice of the two-dimensional transform, F (u, v), subtending
an angle θ with the u-axis.
The result of (5.7) can be generalized by considering the (t, s) coordinate system obtained
by rotating the (x, y) coordinate system an angle θ, such that:
[
t
s
]
=
[
cos θ sin θ
− sin θ cos θ
] [
x
y
]
(5.8)
The projection along a line of constant t is:
P (θ, t) =∫ ∞
−∞
f(t, s)ds (5.9)
And its Fourier transform is:
S(θ, w) =∫ ∞
−∞
∫ ∞
−∞
P (θ, t)e−2πiwtdsdt (5.10)
A transformation to the original (x, y) coordinate system, using (5.8), yields:
S(θ, w) =∫ ∞
−∞
∫ ∞
−∞
f(x, y)e2πiw(x cos θ+y sin θ)dxdy (5.11)
Chapter 5. Methods for Tomographical Reconstruction 44
S(θ, w) = F (w, θ) = F (w cos θ, w sin θ) (5.12)
This gives the equality between the Fourier transform of the object function and the
Fourier transform of the projection. The result is that if the F (u, v) function is known
by taking projections at angles θ1, θ2, ..., θk, the function f(x, y) can be reconstructed
by using the inverse Fourier transform.
f(x, y) =∫ ∞
−∞
∫ ∞
−∞
F (u, v)e2πi(ux+vy)dudv (5.13)
However, in practice it is impossible to have an infinite number of projections and thus,
only a finite number of Fourier components will be known. The object function will be,
for −A/2 < x < A/2 and −A/2 < y < A/2:
f(x, y) ≈1
A2
N/2∑
m=−N/2
N/2∑
n=−N/2
F(m
A,nA
)
e2πi((m/A)x+(n/A)y) (5.14)
As it is impossible to have an infinite number of projections, the function F (u, v) is
known only in radial lines. An interpolation is necessary to reconstruct the object
function, but this does not always lead to stable solutions. It would be better to find
this function by a nearest-neighbor scheme, but radial lines get sparser as the radius
increases, which introduces errors that compromise the quality of the reconstruction.
5.1.3 The Filtered Back Projection
The FBP algorithm is an alternative approach to find the object function, without
direclty performing an inverse Fourier transform on the Fourier components. The idea
is to measure the projections, find their Fourier transform, multiply it by a weighting
function and sum the inverse Fourier transforms of the filtered projections.
The object function of (5.13) can be rewritten using a change of coordinates from the
rectangular (u, v) system to the polar coordinate system (w, θ).
f(x, y) =∫ 2π
0
∫ ∞
0F (w, θ)e2πiw(x cos θ+y sin θ)wdwdθ (5.15)
By splitting the integral over θ in 0 to π and π to 2π, writing t = x cos θ + y sin θ and
using the property F (w, θ + π) = F (−w, θ), the object function may be wirtten as:
f(x, y) =∫ π
0
[∫ ∞
−∞
F (w, θ)|w|e2πiwtdw]
dθ (5.16)
Chapter 5. Methods for Tomographical Reconstruction 45
By the Fourier Slice Theorem, this is equal to:
f(x, y) =∫ π
0
[∫ ∞
−∞
S(θ, w)|w|e2πiwtdw]
dθ (5.17)
Which can be written as:
f(x, y) =∫ π
0Q(θ, x cos θ + y sin θ)dθ (5.18)
with
Q(θ, t) =∫ ∞
−∞
S(θ, w)|w|e2πiwtdw (5.19)
The Q(θ, t) is the filtered projection in which the filtering is given by |w|. The object
function is obtained by summing the filtered projection at different angles. This process
is often referred to as back projecting the filtered function Q(θ, t).
The integration over frequency in (5.19) has to be done over all frequencies. However,
the contribution to the function over a certain frquency is negligible. If W is a frequency
higher than the highest frequency in a projection and the projection is sampled at
intervals of 1/2W , the approximate Fourier transform of the projection is given by:
S(θ, w) ≈ S(
m2WN
)
=1
2W
N/2−1∑
k=−N/2
P(
θ,k
2W
)
e−2πi(mk/N) (5.20)
with N the number of samples of the projection. In that way, the filtered projection for
a discrete number of projections is given by:
Q(
θ,k
2W
)
≈2WN
N/2∑
m=−N/2
S(
θ, m2WN
)∣
∣
∣
∣
m2WN
∣
∣
∣
∣
e2πi(mk/N) (5.21)
for k = −N/2, ...,−1, 0, 1, ..., N/2. The filtered projection can be multiplied by a func-
tion (e.g. Hamming window) to reduce the noise in the reconstructed images. The
convolution theorem can be used to write (5.21) as:
Q(
θ,k
2W
)
≈2WN
P(
θ,k
2W
)
∗ φ(
k2W
)
(5.22)
where φ (k/2W ) is the inverse discrete Fourier transform of the multiplication of the
function used to reduce the noise and |m(2W/N)|, with m = −N/2, ...,−1, 0, 1, ..., N/2.
Finally, the filtered projection can be back projected to obtain the object function:
f(x, y) =πK
K∑
i=1
Q(θi, x cos θi + y sin θi) (5.23)
Chapter 5. Methods for Tomographical Reconstruction 46
where K is the number of projections.
It is important to say that this process is only valid for parallel beam geometry. In
practical applications this beam profile is not very common. The calculations have to
be modified to give the correct results with other geometries. This can be done for fan
and cone beam profiles. For more complex geometries it is not possible to modify the
algorithm and other reconstruction metods must be used.
5.2 Iterative Algorithms
Another approach to reconstructing an object from its projections is based on iterative
methods. The Expectation Maximization (EM) is an algorithm that begins with a trial
object and computes its projections. It then compares them with the real projections.
The trial object is modified in such way that the new projections are more similar to
the real ones. This is done until the two projections perfectly match.
The EM algorithm is very demanding in terms of computational resources. In the
mayority of the cases it is faster to use the FBP algorithm. However, a technique
of Ordered Subsets (OS) can help to speed up the process. The OSEM divides each
iteration in subitarations that use a small amount of projections (as low as 3 or 4).[20]
The iterative algorithms are useful when the number of projections is low, the projections
are noisy or the geometry is complex (see Figure 5.2).
��� ���
Figure 5.2: Comparison between FBP reconstruction showing artifacts due to thesmall number of projections (a) and OSEM3 reconstruction (b).[20]
Chapter 6
Tomographical Imaging with
Medipix2
The process of tomographical reconstruction with Medipix2 is discussed in this chapter.
The first section describes the experimental setup used to take the projections of the
sample at different angles. The next section describes the process of reconstruction of
the sample from the projections. Finally, some examples of reconstructed images are
shown along with a discussion of the results.
6.1 Experimental Setup
The setup used for this part of the project was the same as the one used for radiographical
imaging (see Figure 4.1). The only difference was that a sample holder attached to a
step motor was included, as shown in Figure D.2. The step motor provided the precise
rotation of the sample to take radiographies at different angles.
The Mo tube was set to 35 kV and 1 mA, as before. The distance between the 5 mm
collimator and the Medipix2 was 390 mm. The sample was placed at 355 mm from
the collimator. An exposure time of 2.5 seconds was used for the projections as well
as for the calibration of the beam hardening correction. The sample was rotated, via
a LabVIEW program (see Appendix B), after each projection was taken. It took from
five to ten minutes to acquire 200 projections in 360◦.
47
Chapter 6. Tomographical Imaging with Medipix2 48
6.2 Tomographical Reconstruction with Octopus 8.3
The reconstruction of the sample from the projections was done with Octopus 8.3, a
commercial tomography reconstruction package for CT, developed at the Centre for X-
ray Tomography of the Ghent University. A time-limited full version of the Octopus
8.3 and Octopus 3D Viewer was made available for this project by its developers. The
Octopus 8.3 combines high performance algorithms with a user-friendly interface for
parallel, fan, cone and spiral CT.
The Octopus 8.3 receives the projections in 8-bit (or 16-bit) TIFF images. It has a
module for converting RAW images or images in other formats (i.e. BMP, JPEG or
PNG) to the correct format (see Appendix E). The images are then imported and pro-
cessed. They can be cropped to the desired region of interest, spot filtered, normalized
and ring filtered before producing the sinograms. The spot filter and ring filter correct
detector-based defects and the normalization is the same flat field correction with an
open beam image.[22] After the projections are ready, they are reorganized into sino-
grams (see Figure 6.1). A sinogram is the image that results after a Radon transform.
It is the collection of the projections at different angles of one row of pixels.
Figure 6.1: Example of a sinogram from the cone shell.
Once the sinograms are ready, the slices can be reconstructed. The parameters of the
reconstruction can be modified and individual slices can be reconstructed to test them.
When the settings produce the desired results, the whole volume can be reconstructed.
The reconstruction of the total volume took less than 30 seconds, because of the small
size of the images (256 × 256 pixels) The scaling of the visualization can be changed to
keep the relevant information and the stack of slices is saved as TIFF images. The slices
can be visualized and rendered with the Octopus 3D Viewer.
Chapter 6. Tomographical Imaging with Medipix2 49
6.2.1 Three Objects Phantom
The three objects phantom had a wood stick, a wire from a resistance and a piece of
cable embedded in clay (see Figure 6.2a). The phantom was thought to determine if
it was possible to distinguish among the different materials (i.e. wood, metal, plastic
and clay) in a rather simple geometry. The reconstruction was done assuming a parallel
beam profile from 100 projections (see Figure 6.2b for an example of a projection) in
180◦.
��� ���
Figure 6.2: Photograph (a) and STC-corrected projection (b) of the three objectsphantom. The phantom had a wood stick, a piece of cable and the wire of a resistance.The region of interest was the central part of the phantom at a height between 1 and
2 cm.
The reconstructed slice of the three objects phantom (see Figure 6.3a) is from a height
of about 1.5 cm in Figure 6.2a. The two metal wires appear clear in the slice as well
as in the multi-surface render in Figure 6.3b. The wood stick is barely noticeable in
both figures. This is due to the similarity in absorption of wood and clay. The plastic
cover of the cable is impossible to distinguish because of the same reason. Both images
present artifacts from the reconstruction, due to the small amount of projections (i.e.
100 projections) and due to the fact that the sample was only rotated 180◦. The artifacts
also come from the assumption that the beam profile was parallel, when it was actually
conic, as it came from an X-ray tube. These are the reasons of the black lines between
the two wires, that appear in the render as a plane passing through both wires.
Chapter 6. Tomographical Imaging with Medipix2 50
��� ���
Figure 6.3: Reconstructed slice (a) and multi-surface render (b) of the three objectsphantom.
6.2.2 Helicoidal Wire Phantom
The next phantom was a helicoidal wire embedded in wax, as shown in Figure 6.4a.
The objective of this phantom was to test the cone beam reconstruction with a sample
presenting a more complex geometry than the three objects phantom. Therefore, the
reconstruction was done assuming a point source of radiation, using 200 projections (see
Figure 6.4b for an example of a projection) in 360◦.
������
Figure 6.4: Photograph (a) and STC-corrected projection (b) of the helicoidal wirephantom.
Chapter 6. Tomographical Imaging with Medipix2 51
��� ���
Figure 6.5: Reconstructed slice (a) and multi-surface render (b) of the helicoidal wirephantom.
The reconstructed slice (see Figure 6.5a) shows the wire and a less dense material sur-
rounding it. It is inferred that this corresponds to the plastic cover of the wire. A clear
difference in absorption among these two materials and wax is noticeable in the slice,
as well as in the multi-surface render in Figure 6.5b. The green surface in the render
corresponds to the metal wire, while the blue regions indicate where the absorption is
less than that of metal but higher than wax. The helix is perfectly rendered and the
slice is almost artifact-free, which results in a considerable difference with the former
reconstruction. This can be accounted to the double number of projections and to the
geometry used for the reconstruction (cone beam geometry needs projections in 360◦ and
some additional parameters as source-detector distance and source-object distance).
6.2.3 Cone Shell
The cone shell in Figure 6.6a was reconstructed with the same parameters and number
of projections used for the helicoidal wire phantom. The projection acquisition had to
be divided in two, one for the lower part (see Figure 6.6b) and another for the upper part
(see Figure 6.6c), because of the dimensions of the shell. It was important to reconstruct
the whole volume, as it has an interesting internal structure from top to bottom.
The reconstructed slices in Figure 6.7a and 6.7c show a clear distinction of the material
of the shell. However, some artifacts can be seen in both slices. It is known[20] that
analytic reconstruction algorithms are fully correct when noise levels are low and an
infinite number of projections is available. The Octopus software is based on analytic
Chapter 6. Tomographical Imaging with Medipix2 52
algorithms that give rise to these kind of artifacts, which can be reduced with a greater
amount of projections. Despite the artifacts, the multi-surface renders of the lower
(see Figure 6.7b) and upper parts (see Figure 6.7d) of the shell show an almost-perfect
representation of the internal structure of the shell.
���
��� ���
Figure 6.6: Photograph (a), STC-corrected projection of lower part (b) and STC-corrected projection of upper part (c) of the cone shell.
Two phantoms and one object were reconstructed from their projections at different
angles. Different geometries and compositions were used to test the parameters of the
reconstruction with Octopus 8.3. It was shown that, even though the geometry of the
three objects phantom was rather simple, the low amount of projections (i.e. 100 in
180◦) and the parallel beam profile assumption resulted in a low-quality reconstruction
with artifacts that distorted the slices and surface render. On the other hand, the
reconstruction of the helicoidal wire and the cone shell gave better results, because of
the greater amount of projections (i.e. 200 in 360◦) and the cone reconstruction profile
Chapter 6. Tomographical Imaging with Medipix2 53
��� ���
��� ���
Figure 6.7: Reconstructed slice of lower part (a), multi-surface render of lower part(b), reconstructed slice of upper part (c) and multi-surface render of upper part (d) of
the cone shell.
assumption. The combination of the PHYWE X-ray unit, the Medipix2 detector and the
Octopus 8.3 was optimal for tomographical reconstruction of objects with a resolution
close to 1 mm. For better resoultions it is essential to raise the source-detector distance
or to have an X-ray tube with a smaller focus (i.e. micro or nano-focus).
Chapter 7
Final Discussion
The main objective of this project was to obtain tomographical reconstructions of two
phantoms and an object from projections at different angles, taken with Medipix2. A
review on biomedical imaging was presented and the Medipix2 pixel detector was briefly
described, emphasizing the features that make it interesting. After performing an energy
calibration of the detector, the energy window feature of the Medipix2 was used to take
radiographies of a phantom and an image, at different energies. The flat field and beam
hardening correction were implemented to the radiographies and it was shown that the
beam hardening correction is necessary to produce high-quality images.
The Octopus 8.3 software was used to do the reconstruction process in two phantoms
and an object. The size of the focus of the X-ray tube (∼ 1 mm) and the low intensity
of the beam were not a limitation for the quality of the reconstructed images. However,
if higher resolution is needed, it is essential to change the distance between the source
and the detector, or to use an X-ray tube with a smaller focus.
To improve the experimental setup it is important to construct an automated positioning
device for the sample holder. This upgrade is nedded to take radiographies at a large
scale. It is also useful to automate the beam hardening correction and the acquisition of
projections to reduce the exposure time. This project sets the basis for the construction
of a Computed Tomography facility at UniAndes with the Medipix2 pixel detector.
54
Appendix A
Specifications of Step Motor and
Driver
The step motor used for the sample holder was an Applied Motion Products NEMA 14
5014-820.1 This motor has a step angle of 1.8◦, weight of 0.33 lbs, voltage of 3.2 V and
current of 0.35 A. The dimensions of the motor are shown in Figure A.2.
Figure A.1: Photograph of step motor in which the sample holder was mounted.
The motor was controlled with the PDO-2035 driver from Applied Motion Products
(Figure A.3). This controller has to be connected to a 120 V AC outlet. The step motor
is connected to the output of the driver. The input of the driver controls the direction
and steps of the motor. These parameters are controlled via an external source of 5 V.
The Enable input is to control if there is current or not in the motor. The Direction
input controls the direction of rotation of the motor with a logical 0 or 1. The Step
input uses the rise edge of the 5 V pulse to move the motor one step in the direction
indicated by Direction.1All the information about the step motor and driver was taken from the Applied Motion Products
Website http://www.applied-motion.com/products/stepper/index.php
55
Appendix A. Specifications of Step Motor and Driver 56
Figure A.2: Dimensions of step motor.
Appendix A. Specifications of Step Motor and Driver 57
Besides these controls, the current that passes through the motor has to be selected
with the DIP switches. The current that was selected for the whole project was 0.25
A/phase. The torque requirements were not very demanding due to the small weight of
the samples. However, this has to be re-evaluated for larger samples to prevent undesired
rotations or vibration of the specimen. It is also possible to select between full step (1.8◦
with this motor) or half step (0.9◦ with this motor).
�����
���
�������
�����
��
���
��
���
�����
���������� ������ �� �� ���
�����
��������
��
��
��
��
���
�
�
����������
���� ��������
�����
�����
�����
�����
������
�������������
����
���
�����������
�����
�����
�����
�������
�� �
�����
Figure A.3: Dimensions of step motor driver.
Appendix B
External Control of the Step
Motor Driver
Figure B.1: Interface of the LabVIEW program used to control the step motor.
The inputs of the step motor driver were controlled via a National Instruments PCI-
6053 NI-DAQmx connected to a CB-50 I/O Connector Block. A LabVIEW program was
created (see Figure B.2), with the help of Ing. Marco Antonio Gonzalez (UniAndes),
to control the step motor from the same PC in which the Medipix2 acquisition was
controlled. Figure B.1 shows the interface of the program, in which the current in the
motor can be enabled or disabled and the direction and the number of steps can be
selected. The OK button performs the number of steps selected. The Done display
shows the number of steps the motor has rotated in each cycle. It is reinitialized the
next time the OK button is pressed. The program can be modified to count steps until
a dedicated button is pressed to reinitialize the counter.
58
Appendix B. External Control of the Step Motor Driver 59
Figure B.2: LabVIEW program used to control the step motor.
Appendix C
The PHYWE X-ray Unit
The PHYWE X-ray unit (see Figure C.1) was used to perform all the measurements
and image acquisitions in the project1. The X-ray tube is inside a shielded box with a
safety mechanism that prevents opening the unit while the tube is operating. The unit
has the unique feature of three different tubes with anodes made of Cu, Mo and Fe (only
the Cu and Mo tubes were available). It has a goniometer controlled with a step motor
where a diffracting crystal is placed between the tube aperture and the Geiger counter
holder. The goniometer and the tube are controlled with a microprocessor. The unit
can also be controlled remotely from a PC.
Figure C.1: PHYWE X-ray unit showing the possibility of changing the X-ray tube.
1All the information about the PHYWE X-ray unit was taken from the PHYWE Websitehttp://www.phywe.de
60
Appendix C. The PHYWE X-ray Unit 61
The tube voltage can be set between 0 and 35 kV in 0.1 V increments. The current in
the tube can be set between 0 and 1 mA in 0.1 mA increments. The counting time of
the device can be set between 0.5 and 100 seconds in 0.1 seconds increments.
The goniometer was removed from the unit to place the Medipix2 and the sample holder
inside the radiation field, as shown in Figure D.3.
Appendix D
The Medipix2 Support and the
Sample Holder
The Support for the Medipix2 (see Figure D.1) was designed and constructed at UniAn-
des by Luis Carlos Gomez. It was designed to fit in the PHYWE X-ray unit and to
position the Medipix2 in front of the X-ray tube aperture. Four magnets at the bottom
of the support fix it to the floor of the PHYWE X-ray unit. The Medipix2 is secured to
the support with four screws. It has two screws to move the support horizontally and
vertically to place the detector in the correct position with respect to the beam.
Figure D.1: Medipix2 and support inside the PHYWE X-ray unit.
The sample holder (see Figure D.2) was constructed using a precision slider. The step
motor described in Appendix A was attached to the slider. The additional parts were
designed and constructed by Luis Carlos Gomez. The height of the sample holder can
be varied with the screw of the slider. Two magnets at the bottom of the sample holder
fix it to the floor of the PHYWE X-ray unit. The sample holder has to be positioned
62
Appendix D. The Medipix2 Support and the Sample Holder 63
manually in front of the Medipix2, as well as its height to have the sample in the field
of view of the detector.
Figure D.2: Sample holder inside the PHYWE X-ray unit.
Figure D.3 shows the Medipix2 on the support and the sample holder inside the PHYWE
X-ray unit. The Medipix2 and the sample holder are aligned with the X-ray beam.
Figure D.3: Experimental setup for tomography imaging, inside the PHYWE X-rayunit.
Appendix E
Comments on Octopus 8.3
A full version of Octopus 8.3 and the Octopus 3D Viewer was made available for this
project, by it developers. Without this contribution from the Octopus team, the task of
reconstructing the slices from the projections would have been more difficult due to the
time limitations. The support of Dr. Manuel Dierick from the Ghent University was
valuable throughout the project. All the information about the Octopus software can
be found in the website XRayLAB.com
Comment on Input Images
The radiographies taken with Medipix2 and STC-corrected in Pixelman were saved as
text matrices of 256 × 256. This information was imported to Mathematica 6.0 and
exported as 8-bit TIFF files that were used as the input of Octopus 8.3. This method
allowed to set the same contrast level in the stack of projections and to export them in
an automated way.
64
Bibliography
[1] M. Hoheisel. Review of Medical Imaging with Emphasis on X-ray Detectors. Nuclear
Instruments and Methods in Physics Research A 536, pages 215–224, January 2006.
[2] A. M. Cormack. Representation of a Function by Its Line Integrals, with Some
Radiological Applications. Journal of Applied Physics, 34(2722), September 1963.
[3] A. M. Cormack. Representation of a Function by Its Line Integrals, with Some
Radiological Applications II. Journal of Applied Physics, 35(2908), October 1964.
[4] J. D. Enderle, S. M. Blanchard, and J. D. Bronzino. Introduction to Biomedical
Engineering. Elsevier Academic Press, 2005.
[5] D. Griffiths. Introduction to Elementary Particles. John Wiley and Sons, Inc, 1987.
[6] The Power of Molecular Imaging PET. The Academy of Molecular Imaging. URL
http://www.petscan.org/.
[7] Advisory Commitee on Human Radiation Experiments. U. S. Department
of Energy. URL http://hss.energy.gov/healthsafety/ohre/roadmap/achre/
intro_9_5.html.
[8] R. Eisberg and R. Resnick. Fısica Cuantica. Limusa Wiley, 2005.
[9] Environmental Health and Safety. McGill University. URL http://www.mcgill.
ca/ehs/radiation/basics/units/.
[10] P. Sprawls. Radiation Quantities and Units. The Physical Principles of Medi-
cal Imaging. Sprawls Educational Foundation. URL http://www.sprawls.org/
resources/RADQU/.
[11] C. Peterson. The Charge-Coupled Device, or CCD. Journal of Young Investi-
gators, 3(1), March 2001. URL http://www.jyi.org/volumes/volume3/issue1/
features/peterson.html.
65
Bibliography 66
[12] X. Llopart, M. Campbell, D. San Segundo, E. Pernigotti, and R. Dinapoli.
Medipix2, a 64k Pixel Read Out Chip with 55 µm Square Elements Working in
Single Photon Counting Mode. IEEE, 3(1), November 2001.
[13] X. Llopart. Design and Characterization of 64K Pixels Chips Working in Single
Photon Processing Mode. PhD thesis, CERN, 2006.
[14] C. Lebel. Energy Calibration of the Low Threshold of Medipix USB. Universite de
Montreal, 2007.
[15] Medipix in IEAP. Department of Applied Physics and Technology IEAP. URL
http://aladdin.utef.cvut.cz/ofat/index.html.
[16] M. Fierdele, D. Greiffenberg, Et Al. Energy Calibration Measurements of Medipix2.
Nuclear Instruments and Methods in Physics Research A 536, pages 75–79, March
2008.
[17] L. Tlustos, R. Ballabriga, M. Campbell, Et Al. Imaging Properties of the Medipix2
System Exploiting Single and Dual Energy Thresholds. IEEE, pages 2155–2159,
July 2004.
[18] C. Kittel. Introduction to Solid State Physics. John Wiley and Sons, Inc., 1996.
[19] M. J. Berger, J. H. Hubbel, Et Al. XCOM Photon Cross Sections Database. Na-
tional Institute of Standards and Technology, 2005. URL http://physics.nist.
gov/PhysRefData/Xcom/Text/XCOM.html.
[20] J. Jakubek. Data Processing and Image Reconstruction Methods for Pixel Detec-
tors. Nuclear Instruments and Methods in Physics Research A 576, pages 223–234,
February 2007.
[21] A. C. Kak and M. Slaney. Principles of Computerized Tomographic Imaging. IEEE
Press, 1988.
[22] Octopus 8.3 Manual. XRayLAB. Ghent University, 2008.