Key words
magnetic particle imaging - MR imaging - superparamagnetic iron oxide - preclinical
- workflow - co-registration
Introduction
Magnetic particle imaging (MPI) is a new imaging technique. Since its initial description
[1], many technical improvements and various setups of MPI scanners have been realized
[2]
[3]. Three years later the first in vivo MPI mouse measurement combined with a post
mortem magnetic resonance imaging (MRI) examination was reported [4]. Now, with the first commercially available preclinical MPI scanner, combined preclinical
MR-MPI experiments can be performed to evaluate future applications of a clinical
system [5] ([Fig. 1]).
Fig. 1 Preclinical Magnetic Particle Imaging scanner is installed in a shielded cabinet.
The copper coating shielding eliminates disturbances induced by external radio frequency
signals. The scanner has a circular bore (diameter ~12 cm). With the positioning system,
the animal carrier unit, which is fixed to the technical support (TSU), can be pushed
into the bore. All tubes and cables for narcotics, air for the heating system, and
vital sign derivation are attached to the TSU and are piped to the carrier unit.
Abb. 1 Der präklinische Magnetic Particle Imaging Scanner ist in einem abgeschirmten Raum
installiert. Die Kupferbeschichtung verhindert Störungen durch externe Radiofrequenzsignale.
Der Scanner besitzt eine runde Bohrung (Durchmesser ~12 cm). Mittels des Positionierungssystems
kann die Trageeinheit, die an der technischen Unterstützungseinheit (TSU) fixiert
wurde, hineingeschoben werden. An die TSU werden Schläuche für Narkosegas und Heizluft
sowie eine Stromleitung für die Heizung und Lichtleiter für die Übertragung von Vitalzeichen
angeschlossen und an die Trageeinheit weiter geschleust.
MPI is a fast imaging technique and is free of ionizing radiation. One important issue
in MPI development is the indicator/tracer that has to have specific characteristics
to generate an MPI signal [1]. Ferucarbotran, a superparamagnetic iron oxide (SPIO) nanoparticle [6], was developed as an MRI contrast agent for liver-specific uptake. It was shown
early to be appropriate for MPI use as well [1]. While SPIOs are applied in the first place as negative contrast agents in MRI,
they generate positive signals and perform as a tracer in MPI. Since MP images are
background-free, they contain no anatomical information. To generate anatomical information,
MRI measurements were included in the workflow.
The purpose of the present study was to report the first combined use of in vivo MPI
and in vivo MRI after an intravenous bolus injection of ferucarbotran in a mouse model
and to describe the workflow, which has been established for future in vivo applications
of this promising new imaging modality.
Materials and Methods
Animal handling
The examination of five healthy mice was approved by the local committee on animal
protection (Behörde für Gesundheit und Verbraucherschutz, Freie und Hansestadt Hamburg,
Nr. 42/14). Anesthesia was carried out with 1 – 2 % Isoflurane (l-chloro-2.2.2-trifluoroethyl
difluoromethyl ether) and a flow of 0.5 L/min (Vapor 19.3, Dräger, Lübeck, Germany
and Tec 7 Vaporizer, GE Healthcare, Chalfont St Giles, UK). A catheter (inner tube
diameter 0.28 mm, Portex, Smiths Medical International Ltd., USA) was placed in the
tail vein for the ferucarbotran (Resovist®, available only in Japan) injection.
Workflow process
As MPI (Philips/Bruker Preclinical MPI system) delivers no anatomical information,
MRI scans were included in our workflow process (see [Fig. 2]). For the transfer between the MRI and MPI scanners, the mouse was placed on a dedicated
MPI- and MRI-compatible bench (MINERVE, Esternay, France). On the MRI and the MPI
scanner this carrier was connected to a technical support unit that was attached to
the scanners’ positioning systems. The technical support units facilitate connectors
for gas inflow and outflow, interfaces of vital signals, and for a heating system.
Heated air was flowing through the hollow bench to keep the body temperature at 37 °C.
Respiratory signals were processed by monitoring and gating units on the MRI (SA Instruments
Inc., New York, USA) and MPI site (MINERVE, Esternay, France), respectively.
Fig. 2 A schematic timeline of the performed workflow is shown. MRI is applied before and
after the MPI scan. The dynamic MPI was reconstructed at a later time point.
Abb. 2 Schematische Darstellung des zeitlichen Arbeitsablaufes: MRT-Messungen wurden vor
und nach der MPI-Messung durchgeführt. Die Bildrekonstruktion der dynamischen Messserie
fand zu einem späteren Zeitpunkt statt.
MRI scans were performed with a preclinical 7 T MR system (Clinscan 70/30 with syngo
MR B15, Bruker Biospin GmbH, Ettlingen, Germany) before and after the MPI examination.
While pre-injection MR scans at the beginning delivered anatomical images, the final
scan validated the successful injection during the MPI scan and verified that no dislocation
of the mouse occurred. At the MPI site a baseline scan was first performed. Images
were reconstructed to validate that the mouse was MPI signal-free. Then a dynamic
scan was started and after 1.5 minutes ferucarbotran was injected. Since reconstruction
of a complete dynamic series is a time-consuming process, it was carried out at a
later time point. Instead, we repeated the baseline scan and verified whether signal
enhancement in the reconstructed images of this single 3 D volume was visible. Finally
the mouse was brought back to the MR scanner to validate the SPIO injection by detecting
a signal decrease within the liver on the T2-weighted images.
MPI reconstruction was performed using the system function approach [7]. This implies that a separate calibration scan was performed beforehand using the
same MPI parameters and tracer as used for the later mouse examination. Hereby, a
sample tube containing pure ferucarbotran was moved by a robot through the bore of
the scanner ([Fig. 3]). The reconstruction was processed on the MPI console (ParaVision 6.0/MPI, Bruker
BioSpin MRI GmbH). To visualize and extract dynamic information, additional image
processing software (ImageJ, NIH, USA) extended by our own plugins (qMapIt) was used.
Fig. 3 A robot is used to measure system function a. It moves a calibration sample, which is mounted to the green holder b, through the sampled volume. The size of the sample should not be larger than the
voxel resolution of the reconstructed MPI images. Therefore, various sized tubes are
available, in which the tracer has to be filled.
Abb. 3 Zur Messung der Systemfunktion wird ein Roboter verwendet a. Er fährt eine Probe, die an einem grünen Stab befestigt ist b, durch das Messfeld. Die Größe der Probe sollte nicht größer als die Voxelauflösung
der rekonstruierten Bilder sein. Daher stehen verschieden große Probenröhrchen zur
Verfügung, in die das bei der Injektion verwendete Tracer pipettiert wird.
In the final processing step, the images were co-registered (Imalytics, Philips Medical
Systems, Aachen, Germany) and the results were stored on a picture archiving system
(dcm4chee 2.17, dcm4che.org). To guarantee correct co-registration of MPI and MRI
data, the possibility of dislocation of the mouse during the transfer between the
two scanners was minimized. Firstly, an MPI and MRI-compatible carrier unit (Equipement
Vétérinaire MINERVE, Esternay, France) suitable for both systems was used to avoid
rearrangements ([Fig. 4]). Secondly, the time of transport was short. The distance between the MPI and MRI
scanner site was only 10 meters.
Fig. 4 Two technical support units (TSU) are shown. The upper TSU will be attached to the
MPI positioning system. The lower one is used for MRI measurements. It has a mounted
mouse carrier unit which can be locked by a screw. Both TSUs feature connectors and
interfaces for narcotics, air for the heating system, and vital sign derivation.
Abb. 4 Dargestellt sind zwei technische Unterstützungseinheiten (TSU). Die obere TSU wird
an das Positionierungssystem des MPI-Scanners befestigt, die untere wird für MRT-Messungen
vorgesehen. An dieser TSU ist eine Maustrage angesteckt, welche durch eine Schraube
fixiert wird. Die beiden TSUs bieten Anschlüsse und Schnittstellen für die Narkose,
Luft für das Heizsystem und zur Vitalzeichenableitung.
MRI protocol
The MRI protocol consisted of a survey scan and three respiratory-triggered T2-weighted
scans which covered the chest and abdomen in a coronal, sagittal, and transverse orientation.
The following scan parameters were used for the 2 D turbo spin echo sequence: field
of view (FOV) 32 mm, matrix 256 × 256, slices 28, thickness 0.8 mm (no gap), TR 1100 ms
(triggered on every respiratory cycle at 40 cycles per minute), TE 28 ms, turbo factor
8, NSA 3 with a scanning time of 8 minutes each. These images were used as an anatomic
reference when co-registered with MP images.
MPI protocol
All MPI scans were performed with the same hardware settings. The gradient of the
selection field was 1 T/m. This gradient defines the inherent resolution of the field
free point (FFP). The amplitude of the drive field, which moves the FFP, was 10 mT.
Together they define the sampled volume, which was a volume of 40 × 40 × 20 mm³ in
our setting. The MPI procedure was carried out in three steps. First, a baseline control
measurement was performed validating the absence of an MPI signal. The scanning time
with 1000 averages was 21.5 seconds. Then a dynamic scan consisting out of 28 000
repetitions without averaging was performed. It had a scanning duration of 10:03 minutes
and a temporal resolution of 21.5 msec per 3 D volume. After the 3400th repetition,
50 µL of 500 mM ferucarbotran were injected within 4 seconds. While the injection
was performed manually in the first mouse, later experiments were performed with a
syringe pump (AL1000 – 220Z, World Precision Instruments, Berlin, Germany) steered
by the scanner’s software.
MPI system function, reconstruction and image co-registration
For the system function measurement, a point sample (Bruker Biospin GmbH, Ettlingen,
Germany) filled with 7.8 µL ferucarbotran covering a squared area of 2.5 × 2.5 mm²
was used. The system matrix consisted of 20 × 20 × 20 voxels covering a field of view
of 60 × 60 × 30 mm³. It was scanned with 100 averages at each position within 9:16
hours. Using this system function, images were reconstructed using the following parameters:
Kaczmarz algorithm, iterations 15, regularization parameter 10–4, bandwidth 0.08 – 0.625 MHz, signal-to-noise ratio threshold 1.4 (see [8] for details of MPI reconstruction). MPI raw data can be averaged to improve signal
quality. This also reduces the reconstruction time of a whole dynamic series. We reconstructed
3 D MP images in blocks of 25 and of 100 sequentially acquired raw data sets. This
resulted in a sampling time rate of 538 milliseconds and 2.15 seconds, respectively.
The 3 D volume with 100 averages of the time frame shortly after injection was then
co-registered with the MR images by mutual alignment using a rigid transformation.
The window settings were adjusted to highlight areas of maximal signal intensity.
Results
The combined MR-MPI measurements were successfully carried out in all five mouse experiments
without dislocating the mice between MRI and MPI. The comparison between pre- and
post-injection MPI gave an indication that the injection was successful. This was
confirmed by MRI when comparing the pre- and post-injection images, as the expected
signal decrease in the liver was observed ([Fig. 5]). In the first step, the interpretation of single MP images was hindered since spatial
resolution is low and anatomic references were missing. A maximum intensity projection
over time frames and interpolation to a higher image matrix was used to generate an
overview ([Fig. 6]). The analysis of the dynamic signal changes in depicted regions with a high MPI
signal revealed distinct injection signal intensity time curves ([Fig. 7]). After co-registration with the high-resolution MR data, these areas could be matched
with the cavities of the inferior vena cava und the heart ([Fig. 8]). Thereby, we iteratively repeated the adjustments of the rigid transformation on
all three orientations. In the sagittal view, we observed a hyperintense signal region
in the chest, which could be allocated in the coronal view to the heart. In the sagittal
orientation the long stretched structure starting from the heart following caudal
along the spine could be allocated to the inferior caval vein. The axial view through
the liver (not shown) and the kidneys ([Fig. 8]) proved the signal origin as the vena cava. A coronal image through the kidney proved
this as well as sagittal images.
Fig. 5 Coronal T2-weighted MR images of the mouse trunk before and after an injection of
the SPIO Ferucarbotran. Ferucarbotran significantly decreases the liver signal due
to the uptake in liver macrophages (Kupffer cells). No displacement of the mouse between
the MRI measurements could be observed.
Abb. 5 Dargestellt ist eine koronare T2-gewichtete MRT des Mausstamms vor und nach Injektion
von Ferucarbotran. Ferucarbotran führt zu einer deutlichen Signalminderung in der
Leber durch Aufnahme in Lebermakrophagen (Kupffer-Zellen). Es trat keine Verschiebung
zwischen den MRT-Untersuchungen auf.
Fig. 6 MPI images of ten coronal slices a without and b with bilinear interpolation. Images were obtained after maximum intensity projection
of the time series.
Abb. 6 MPI-Bilder von zehn koronaren Schichten a ohne und b mit bilinearer Interpolation. Die Bilder wurden mittels maximaler Intensitätsprojektion
über die Zeitserie berechnet.
Fig. 7 Two coronal slices a out of a 3 D MPI data set at the time point of tracer injection. The image matrix
is 20 × 20 pixels and covers an area of 6 × 6 cm². Two regions are marked red and
blue that demonstrate b a significant perfusion peak immediately after injection. By co-registering with
MR data, the regions could be assigned to the inferior vena cava and the heart, respectively.
Abb. 7 Zwei koronar-orientierte Schichten a aus einem 3D-MPI-Datensatzes während der Tracerinjektion. Die Bildmatrix ist 20 × 20
Pixel groß und deckt einen Bereich von 6 × 6 cm² ab. Rot und blau markiert sind zwei
Regionen, die b ein deutliches dynamisches zeitliches Verhalten wiedergeben. Gut erkennbar sind die
ausgeprägten Perfusionspeaks. Mittels einer Co-Registrierung mit MRT-Datensätzen konnten
die Regionen der unteren Hohlvene und dem Herzen zugeordnet werden.
Fig. 8 While MRI is providing the anatomic information, the MPI signal shows the inflow
of the tracer. a Sagittal view of co-registered MRI and MPI data. Further examined intersection planes
are marked in blue and orange lines. b In the coronal orientation the MPI signal can be clearly allocated in the heart cavities.
c The MPI signals can be allocated in the sagittal (solid blue), coronal (fine dashed
blue) and transverse (dashed blue) images in accordance with the inferior vena cava.
Abb. 8 Während das MRT die anatomische Information liefert, gibt das MPI-Signal die Traceranflutung
wieder. a Sagittale Darstellung der Co-Registrierung von MRT- und MPI-Aufnahmen. Weitere untersuchte
Schnittebenen sind durch blaue und orange Linien gekennzeichnet. b In koronarer Schichtführung kann das MPI-Signal eindeutig in den Herzhöhlen lokalisiert
werden. c Die MPI-Signale können in der sagittalen (blau durchgezogen), der koronaren (blau
fein gestrichelt) und der transversalen (blau gestrichelt) Schnittebene in guter Übereinstimmung
der unteren Hohlvene zugeordnet werden.
Discussion
With this study we present the first in vivo MPI images acquired during a complete
in vivo MR-MPI workflow with the first commercially available preclinical MPI scanner.
We repeated the workflow in five mice. Images shown in the publication are derived
from the first examined mouse. The inferior vena cava was imaged sufficiently. This
is very promising for future applications like magnetic particle angiography and MPI
perfusion imaging since the scanner´s diameter allows also the examination of larger
animals with larger vessels. Our measurements demonstrate the need of a combined in
vivo MR–MPI workflow. In our case, we performed sequential measurements using MRI
before and after the MPI scan. Concepts of combined MR-MPI systems already exist and
prototypes are under development [9]. A combined system would facilitate the positioning and the image co-registration
process. Positioning in MPI without a reference bears the risk of missing the volume
of interest. We used a drive field of 10 mT and a gradient of 1 T/m to cover two centimeters
in the scanners’ z-direction, which matched the diameter of the mouse. The FOV size
scales linearly with the drive field amplitude. Furthermore, an increase in the drive
field amplitude might affect the signal-to-noise ratio (SNR) positively [2].
A possible solution to avoid misplacement of the FOV would be use of fiducial markers
[10]. Fiducial markers detectable in both modalities, here MPI and MRI, were not available
at the time of our experiment. They might not only be used for the purpose of locating
the volume of interest but also to support image co-registration. The need of an anatomic
reference is comparable with PET or SPECT imaging. The images matched well after use
of a rigid transformation, which takes translation and rotation into consideration.
Improving resolution is an important topic when visualizing vessels. The resolution
limit depends on the magnetic properties of the tracer and the selection field strength
[2]. In contrast, the reconstructed voxel resolution is defined by the field of view
and the sampling step size of the system function matrix. For the applied setup we
achieved a voxel size of 3.0 × 3.0 × 1.5 mm³. Still, we were able to detect the vena
cava shortly after the bolus injection, which had a diameter of only 1.2 – 1.6 mm.
This is very promising for the development of a clinical scanner and its achievable
resolution and underlines the potential use of MPI for angiographic purposes. The
use of technically possible 2.5 T/m instead of 1.0 T/m would have improved resolution
but would have decreased the size of the FOV. The previously reported prototype scanner
had a selection field gradient strength of a maximum of 5.5 T/m offering submillimeter
resolutions [4]
[11]
[12]. However, it must be noted that its inner bore diameter was 32 mm in contrast to
the diameter of 119 mm of our system. Larger bore sizes increase hardware requirements
enormously.
As pointed out earlier, the ratio between drive field amplitude and gradient strength
defines the dimensions of the FOV. To keep the FOV constant we would have to use 25
mT at 2.5 T/m while 12 mT was the current preliminary scanner limit. Further improvements
of the drive field strength are on the way. They will increase the flexibility of
the system and its general performance. On the other hand, a large FOV and high resolution
require system function that has to be more densely sampled using smaller steps of
the robot and a wide range of positions. This means a drastically longer sampling
time for the calibration scan and higher memory load for the computer. Possible solutions
are already being discussed such as simulated system functions [13] instead of measured ones, or strategies to undersample calibration scans [14].
Sensitivity is a key aspect, but was not addressed in our study. We injected a bolus
of 50 µL ferucarbotran to generate an MPI signal that is 30 times the noise background
when the reconstruction was performed with an averaging block size of 100 repetitions
([Fig. 7]). When the data was reconstructed with a smaller block size of 25 repetitions, the
enhancement ratio decreased and the curve became noisier. The applied volume of 50 µL
corresponds for instance in a 28 g mouse to a dose of 0.9 mmol/kg or an iron blood
concentration of about 0.7 g/L. Ferucarbotran was applied in humans to detect local
liver lesions. Generally, for patients above 60 kg, 1.4 mL of a 500 mM solution were
injected, which relates to a dose of 11.7 µmol/kg or an iron blood concentration of
approximately 8.1 mg/L. The applied MPI dose was significantly higher also in comparison
to the formerly reported first in vivo MPI measurement [4]. Here the hardware specifications were optimized for mouse examinations, but our
scanner is equipped with large receive coils and a larger bore diameter and can also
be used for larger animals at the expense of a sensitivity loss for smaller animals.
Similar to MRI, dedicated receive coils are beneficial from the SNR point of view.
From our data we conclude that the dose can be reduced by a factor of three detecting
a signal that is distinguishable from noise. In the near future, we will repeat the
measurement to determine thresholds regarding the amount of contrast agent that is
sufficient for decent image quality. Then the resolution will also be improved by
combining higher drive field amplitudes and gradient strength. Further development
of MPI contrast agents promises to generate stronger signals and to improve spatial
resolution [15]. This is also necessary due to the fact that Resovist® is commercially available only in Japan. Finally, the process of image reconstruction
has to be improved. Reconstruction is time-consuming and the quality depends on the
choice of reconstruction parameters. The optimization of these parameters is performed
on a subjective scale. Tools for an objective evaluation do not exist yet and have
to be developed. The reconstruction process itself can be accelerated by applying
compression techniques [16].
We conclude that with this first experiment a combined workflow process between a
preclinical MRI system and the first commercially available preclinical MPI scanner
has been demonstrated to be feasible. More work has to be done to ease the workflow
process. A major drawback is the time-consuming reconstruction procedure.
