Spectroscopic biomedical imaging with the Medipix2 detector.

Abstract

This study confirms that the Medipix2 x-ray detector enables spectroscopic bio-medical plain radiography. We show that the detector has the potential to provide new, useful information beyond the limited spectroscopic information of modem dual-energy computed tomography (CT) scanners. Full spectroscopic 3D-imaging is likely to be the next major technological advance in computed tomography, moving the modality towards molecular imaging applications. This paper focuses on the enabling technology which allows spectroscopic data collection and why this information is useful. In this preliminary study we acquired the first spectroscopic images of human tissue and other biological samples obtained using the Medipix2 detector. The images presented here include the clear resolution of the 1.4mm long distal phalanx of a 20-week-old miscarried foetus, showing clear energy-dependent variations. The opportunities for further research using the forthcoming Medipix3 detector are discussed and a prototype spectroscopic CT scanner (MARS, Medipix All Resolution System) is briefly described.

Full text

Page 1

Australasian Physical & Engineering Sciences in Medicine Volume 31 Number 4, 2008

300

Spectroscopic biomedical imaging with the
Medipix2 detector



T. R. Melzer1,2, N. J. Cook3, A. P. Butler1,2,4,5, R. Watts1, N. Anderson2, R. Tipples3 and P. H. Butler1,5


1Physics and Astronomy, University of Canterbury, Christchurch, New Zealand
2Christchurch Medical School, University of Otago, Christchurch, New Zealand
3Canterbury District Health Board, Christchurch, New Zealand
4Electrical and Computer Engineering, University of Canterbury, Christchurch, New Zealand
5European Organisation for Nuclear Research, Geneva, Switzerland


Abstract
This study confirms that the Medipix2 x-ray detector enables spectroscopic bio-medical plain radiography. We show that
the detector has the potential to provide new, useful information beyond the limited spectroscopic information of modern
dual-energy computed tomography (CT) scanners. Full spectroscopic 3D-imaging is likely to be the next major
technological advance in computed tomography, moving the modality towards molecular imaging applications. This
paper focuses on the enabling technology which allows spectroscopic data collection and why this information is useful.
In this preliminary study we acquired the first spectroscopic images of human tissue and other biological samples
obtained using the Medipix2 detector. The images presented here include the clear resolution of the 1.4mm long distal
phalanx of a 20 week old miscarried foetus, showing clear energy-dependent variations. The opportunities for further
research using the forthcoming Medipix3 detector are discussed and a prototype spectroscopic CT scanner (MARS,
Medipix All Resolution System) is briefly described.


Key words Medipix, spectroscopic x-ray imaging,
single photon counting


Introduction

Dual-energy and spectroscopic CT
Until recently, medical x-ray imaging has lacked the
ability to discriminate between different photon energies1.
With this study, our group becomes the first to acquire 2D
spectroscopic images of human tissue using the Medipix2
detector2. This is an important first step in enabling the
development of a practical spectroscopic Computed
Tomography (CT) system.
CT has become an indispensable component of
virtually every radiology department1,3; improvements in
speed and resolution have seen it emerge as the workhorse
for many radiological studies. Despite its usefulness,
current CT detector systems waste much of the information
carried by incident photons. Most CT advancements
have been to improve the use of absorption information
and to reject scattering information. However diffraction,
Corresponding author: Anthony P. Butler, Physics and
Astronomy, University of Canterbury, Christchurch, New Zealand
Tel: (+64) 3 3640913, Fax: (+64) 3 3640620
Email: anthony@butler.co.nz
Received: 29 September 2008; Accepted: 8 December 2008
Copyright © 2008 ACPSEM
interference, and energy information have remained
undetectable and untapped except in the recently developed
dual energy x-ray CT systems1.
Of all the areas of possible improvement for radiation
detectors, energy discrimination of incident x-ray photons
is the most promising1,4-6. The first steps along the path
toward spectroscopic CT have already been taken by dual-
energy CT. This technology has demonstrated that, even
with the relatively simple information yielded by
overlapping spectra, there are significant improvements to
be made to image quality and content.
Digital subtraction of dual-energy CT images uses
differences in attenuation to identify different tissues or
contrast agents3,7-8. The method requires two scans at
different energies, which are digitally subtracted to exploit
different tissue’s energy-dependent absorptions. The
resulting image information can be used to diagnose a wide
variety of abnormalities including the difference between
benign and malignant growths9, calcifications including
microcalcifications in mammography, calcified pulmonary
nodules, and calcified lymph nodes not visible in standard
projection radiographs8,10. Dual-energy subtraction also
assists in the identification of mediastinal masses, the
quantification of vascular disease, recognition of bone,
pleural, and chest wall abnormalities, and the localization
of devices such as stents and catheters which are often
difficult to distinguish using plain projection radiographs11.
Dual-energy CT systems have been developed using X-rays

Page 2

Australas. Phys. Eng. Sci. Med. Vol. 31, No 4, 2008 Melzer et al • Spectroscopic biomedical imaging with Medipix2

from a single source (fast kV switching) as well as a dual-
source1 (X-ray tubes at 90º). They are useful for the above
reasons but do exhibit time lag and count rate limitations;
they are also limited to two energies with overlapping x-ray
spectra.
Spectroscopic CT would significantly enhance this
already useful technology4-5 by dividing a single, broad X-
ray spectrum into definable and separate energy bands.
Allocating the energy of each detected photon to one of
these bands will increase tissue discrimination and alleviate
many current imaging problems with consequent
improvement to imaging practices12. Others have
acknowledged1,6 the expected benefits of energy sensitive
CT and are actively pursuing different methods of attaining
this goal; to date their published efforts have not resulted in
a practical detector technology.
It is difficult to predict the exact consequences that
spectroscopic capabilities will have on x-ray imaging, but
numerous fields within CT are poised to make considerable
advancement with this technology13-15. For example, beam
hardening causes characteristic artefacts in reconstructed
images when areas of low attenuation are falsely computed
around areas of high attenuating materials4-5,12. Materials
such as metallic implants distort the area just around the
implant, which often corresponds to the region of clinical
interest4,12. Current correction methods (filtering or
software algorithms) are imperfect, but energy information
may completely eliminate these artefacts1,5 by measuring
the transmitting spectrum then compensating for the
hardening effect.
A spectroscopic CT scanner would also reduce the dose
required to obtain contrast-dependent images; images from
existing multi-phase contrast protocols could be
reconstructed from a single data acquisition. With energy
resolving detectors, the contrast agent could be selected and
viewed or ignored, facilitating superior identification of
tumours or plaques with decreased dose14. With a single
acquisition, patient movement becomes a non-issue and
productivity increases. Less contrast agent is required,
providing benefits for the elderly and patients with renal
failure or diabetes. These factors translate to safer, as well
as cheaper, CT imaging practices4,12,15.
Spectroscopic information will also yield better
intrinsic tissue contrast4-5. Different tissues have different,
energy-dependent attenuation properties depending on their
atomic makeup. Current dual-energy techniques exploit
these differences to identify specific tissues; full
spectroscopic imaging would enhance this ability3-5,9,16-17.

Dual-energy Medipix2 imaging
In this preliminary study, we focus on demonstrating
the practical application of the Medipix2 detector to plain,
dual-energy imaging of bio-medical samples. Future work
will apply the principles developed here to 3D imaging
where the removal of overlying tissues markedly increases
the power of the technique.
CERN (Conseil Européen pour la Recherche Nucléaire)
designed and developed the Medipix2 detector18. The
photon counting pixel detector uses a semiconductor sensor
layer to directly convert x-ray photons into electron-hole

301
pairs. Under the influence of a bias voltage, the electrons or
holes are collected by a separate electronics layer where the
signal processing occurs. The charge pulses created by the
electrons or holes are individually analysed and counted.
The pixelated sensor and electronics layers allow for real
time signal processing and in-pixel digitising19. These
features of Medipix2 allow for good spatial and temporal
resolution, as well as very low noise2,20.
One of the most exciting capabilities of the Medipix2
chip is its ability to discriminate among different photon
energies, as demonstrated by Tlustos et al 20. The energy
window feature of the Medipix2 chip can select specific
photon energies from the range of energies produced by a
broad bremsstrahlung source. The detector has two pulse
height discriminators, which can be configured to capture
images in a specific energy band. In this manner, two
images are acquired from the same source, one at high
energy and the other at low energy which, when subtracted,
reveal absorption differences. The planned Medipix3
detector described by Ballabriga et al18 has multiple energy
windows facilitating simultaneous capture of images in
several energy bands giving single acquisition, full
spectroscopic imaging.
The technical features of the Medipix2 detector—its
energy discrimination capabilities,
resolution, low noise, and fast readout—make the detector a
good candidate for a prototype spectroscopic CT
system20,21. Such a system has been designed and built by
our group. It has been named the Medipix All Resolution
System (MARS) and we will present full spectroscopic, 3D
images from this system in future publications.



Materials and methods

The work for this paper was carried out on a
decommissioned CTG Senographe 600T (General Electric
Medical Systems, WI, USA) and a Siemens 3000 Nova
Mammomat (Siemens Medical Solutions, Ltd., Erlangen,
Germany) machine still in clinical use. This project was
approved by the local Christchurch School of Medicine
Ethics Committee (URB/07/02/001).
Medipix2 (Figure 1) is a complementary metal-oxide
semiconductor with single photon counting capabilities2.
The detection system contains a matrix of 256 × 256 pixels.
Each 55 × 55 μm2 pixel contains a low noise preamplifier
and two pulse height discriminators. The readout chip is
bump bonded to a 300 μm silicon semi-conducting sensor
layer for direct charge conversion of incoming photons.
This type of hybrid chip separates the detector layer from
the electronics layer allowing the use of different detector
materials. The chip is read out serially by the MUROS2
(Medipix2 re-Usable Read-Out System version 2)
electronic interface controlled by the software package
Pixelman (Czech Technical University, Prague)22.
One of the issues with hybrid detectors is the charge
sharing which occurs when multiple pixels register hits
from a single incident photon interaction with the
semiconducting layer2,22-23. The charge cloud produced by
superior spatial

Page 3

Australas. Phys. Eng. Sci. Med. Vol. 31, No 4, 2008 Melzer et al • Spectroscopic biomedical imaging with Medipix2


302


Figure 1. Photo of the Medipix2 detector with the semiconductor
sensor layer bump bonded to the read-out chip.




Figure 2. I-125 Spectrum: Counts vs. THL. The spectrum of an I-
125 source with charge sharing effects present (non-zero counts
outside the peak). The peak (27.5keV) corresponds to a THL value
of approximately 181.

The incident x-ray photon spreads out as it travels
through the semiconductor under the influence of the bias
voltage. Upon reaching the electronics layer, the charge
cloud can cover many pixels, each of which records a hit
of some fraction of the original photon energy. This sharing
of charge degrades the system’s energy and spatial
resolution, especially at lower energies2,22. Increasing the
bias voltage reduces this effect by decreasing the time for
charge to spread among adjacent pixels but cannot
eliminate it.

Chip calibration
The threshold equalisation process corrects for
inhomogeneities in the pixel response due to variations in
the electronics layer, particularly the variation of
preamplifier gain19. Every pixel contains a 3-bit threshold
adjustment that can be manipulated to equalize that pixel’s
response. We created the low threshold equalisation
distribution by following a method outlined in Bisgoni et
al24 and Tlustos et al20 which, when applied across the
detector array, compensates for electronic inconsistencies
and variation, giving a flatter response.
Detector function is controlled by a number of Digital
to Analogue Converters (DACs). The detector functions in
single energy threshold mode if the THH (High Threshold
discriminator) DAC is set lower than the THL (Low
Threshold discriminator) DAC. In this configuration all
photons with energies greater than THL are recorded. If the
THH is set above the THL, the detector will perform in
energy window mode and will only record photons with
energies between the low and high thresholds.
The high threshold equalisation distribution is
calculated after the determination of a suitable low
threshold distribution. The low threshold is calculated using
the noise floor of the chip as a reference but the high
threshold is calculated using the crossing point of the low
and high threshold distributions20. This means that the high
threshold distribution can never be as accurate as the low
since it will be broadened by the low energy distribution.
The images discussed later in this paper were therefore
acquired using the detector in the single, low energy
threshold mode.

Photon energy scale calibration
Imaging an Iodine-125 seed (peak emission 27.5keV)
permitted the calibration of the threshold DACs (THL and
THH) to x-ray photon energies. This makes the
investigation of known energy windows possible. Scanning
in one DAC increments from the noise floor of THL 155
(THH 0) to THL 200, the detector recorded the cumulative
signal deposited by the seed. Then, following the method
outlined in Tlustos et al20, the count rates from individual
pixels were differentiated with respect to energy and
normalized, yielding the iodine spectrum (Figure 2). THL
~ 181 therefore corresponds to 27.5 keV and the noise floor
usually falls between 2 – 5 keV19. Bisgoni et al24
demonstrated a linear relationship between DAC values and
photon energy; therefore, an approximate energy scale can
be assigned to the THL values using the peak energy of
iodine and the assumed energy of the noise floor. One unit
on the DAC scale corresponds to a photon energy of
between 1.02 and 1.16 keV.

Beam hardening
The silicon detector pixels exhibit an energy-dependant
absorption efficiency which leads to increased image noise
due to beam hardening. The Pixelman software has a
routine to compensate for this effect by obtaining flat-field
images for varying degrees of beam hardening and plotting
the response of each pixel25. The efficiency of each pixel
can therefore be decoupled from photon energy and the
noise due to beam hardening is reduced. This calibration
corrects for the variation in each pixel’s response to
different energies but retains tissue dependent absorption
information22.

Biological sample image acquisition
As an initial step, we imaged sections of chicken wing
to confirm that the detector and calibrations were working
properly. The CTG Senographe 600T was used to capture

Page 4

Australas. Phys. Eng. Sci. Med. Vol. 31, No 4, 2008 Melzer et al • Spectroscopic biomedical imaging with Medipix2


303

14 mm

(a)





calibration appropriate to each THL value was applied to
each image and dead pixel columns were linearly
interpolated. We employed Matlab (The Mathworks,
Massachusetts, USA) to digitally subtract the single
threshold images (Figure 3c). The greyscale window and
level values were adjusted to display maximum detail in
each image.
We also acquired images of the left hand of a 20-
week-old miscarried fetus (Siemens 3000 Nova
Mammomat at 35kVp). Single threshold images were
acquired for the same THL values (155 and 165) and the
exposure times were adjusted as before (Figure 4a, 4b).
The same process of beam hardening compensation and
digital subtraction was performed (Figure 4c).


Results

Projection and dual-energy subtracted images
demonstrate clinical quality over the 14mm x 14mm field
of view. Figure 3 shows the single threshold projection
images of the chicken wing. Taken at the lowest
threshold above the noise floor (THL=155), Figure 3a
(b)
14 mm

(c)






Figure 3. Images of the tip of a chicken wing; all the images
have an area of 14 x 14 mm2. (a) The image was acquired at
THL 155 (E>2.5keV), just above the noise floor. (b) The same
image taken at THL 165 (E>13.4keV). Soft tissue detail is
superior in (a) due to the contribution of the lower energy
photons. These low energy photons are rejected at the higher
THL in (b). Bony detail is preserved in both images. (c) Dual-
energy subtraction image of a chicken wing. THL 165 – 155:
(E>13.4keV – E>2.5keV). The subtraction highlights soft tissue
contrast while removing bone information. In the clinical
environment, this technique helps differentiate between benign
and malignant tissues.


exhibits both bony detail as well as soft tissue
information. Figure 3b rejects these lower energy photons
and subsequently displays slightly less soft tissue
contrast.
Digitally subtracting two images obtained at different
single thresholds produces a dual-energy subtraction
image. Between the two images, the absorption of bone
varies little, while soft tissue attenuation changes
significantly. Upon subtraction, the unchanging bone is
removed and the subtle variations in soft tissue absorption
are obvious upon visual inspection (Figure 3c).
Figure 4 shows the left hand of a 20-week-old
miscarried fetus. The Medipix2 system clearly images the
1.4 mm long distal phalanx in the upper left hand corner
of the image. Figures 4a and 4b are single-threshold
projection images acquired at just above the noise floor
(THL 155) and at THL 165 respectively. The images
exhibit good spatial resolution, high contrast, and low
noise. The digitally subtracted image in Figure 4c
demonstrates the ability to discriminate among different
photon energies and shows the absorption variation due to
different stages of calcification of the bones.

Page 5

Australas. Phys. Eng. Sci. Med. Vol. 31, No 4, 2008 Melzer et al • Spectroscopic biomedical imaging with Medipix2


304
14 mm

(a)

(b)
14 mm

(c)


Figure 4. Fingers of a fetal left hand. (a) THL 155: (E > 2.5keV)
The images display very good spatial resolution. The small,
developing bones and soft tissue of the hand can be seen quite
clearly. The small bone in the upper left hand corner measures
less than 1.4mm in length. (b) THL 165: (E>13.4keV) Eliminating
low energy photons demonstrates slight differences in absorption
that can be seen in the soft tissue and at the edges of bone. (c)
THL 165 – 155: The subtraction image emphasises the different
tissues present. The added amount of material at the edge of the
growing bone causes a sharp increase in contrast and can be seen
clearly in the image. The soft tissue of the fingers, in contrast, is
uniform. The sharp change in contrast highlighted by the
subtraction can be very useful in the identification of
microcalcifications in mammography.




Discussion

The clinical images produced in this study exhibit two
exciting features of the Medipix2 detector. The detector’s
ability to set energy thresholds allows for energy
discrimination and spectroscopic imaging through dual-
energy digital subtraction. Secondly, the radiological
images demonstrate high quality spatial resolution, low
noise, and good contrast.
We obtained both high and low energy images using
the adjustable energy threshold at a constant tube kVp.
Current technologies require a dual-energy source with two
distinct tube kVp’s, or a sandwich setup with two
separately filtered detectors to produce images required for
subtraction. Medipix2 uses a constant kVp bremsstrahlung
source and selects the energy ranges digitally.
The 27.5keV gamma emission peak of Iodine-125
approximately corresponds to THL value of 181. Another
one or more precisely defined reference points (e.g.
gamma emissions) would facilitate investigation of
specific photon energies. In the future, k-emission lines
from irradiated metal foils will be used for such a
calibration.


The silicon semiconductor layer has very limited
absorption efficiency at diagnostic X-ray energies.
Improved production processes will make available higher
z detector materials such as GaAs and CdTe, increasing the
functionality of Medipix by expanding its useful range to
higher, diagnostic energies (>100keV)23,26.
The small size of the chip (14 x 14 mm2) limits its use
in everyday clinical application. For general plain
radiologic imaging, the chip would need to cover a much
larger area. The most promising application at present is to
include a line of detectors in a computed tomography
system. This technique eliminates the need for a large
detector and provides the added benefit of 3-D data sets.
Exploiting this solution, our group has designed and built a
desktop, spectroscopic CT scanner (MARS) to further
explore the capabilities of Medipix2, as well as the
forthcoming Medipix3. The next generation chip,
Medipix3, will have 8 independent threshold values which
will permit full spectroscopic imaging with a single
exposure. Additionally, neighbouring
communicate with one another on an event by event basis,
to reduce the effects of charge sharing. These advances will
create true, spectroscopic images with improved energy and
pixels will