|
Material
and Methods
An
ECAT HR+ PET system (Siemens CTI Co., Erlangen, Germany) is
available for PET patient studies. The system provides 63
slices within a 15.5 cm axial field of view. There are 576
detector crystals per ring with a crystal size of 4.39x4.05x30
mm. The 82944 lines of response per plane are usually reduced
by a standard angular compression factor of two. Typically,
23 frames are acquired for 60 minutes following the intravenous
injection of F-18-deoxyglucose (FDG). A total of 1449 cross
sections (23 frames x 63 cross sections) are reconstructed
from one dynamic series. Besides a dynamic acquisition, static
acquisitions are usually performed at 1-3 additional bed positions.
Generally, transmission measurements (10 min for the dynamic
series, 5 min for each additional static acquisition) preceded
all emission acquisitions.
A subnet
of PC systems running Windows 2000 professional server and
Windows 2000 professional (Microsoft Co., Redmond, USA) are
used for PET data processing.
Fig 1a
click
= zoom
Currently
three double processor systems and ten single processor computers
are available within the PET subnet and used for image reconstruction
as well as qualitative and quantitative data evaluation. The
acquisition data are transferred from the PET system to the
subnet server using the file transfer protocol (ftp). The
program for the iterative image reconstruction is written
in C (Visual C++ 6.0, Microsoft Co., Redmond, USA) and is
running as a background job on PC systems within the PET subnet
(6). Each active reconstruction program is checking the subnet
server for new reconstruction tasks at a one minute interval.
The reconstruction parameters are provided using a javascript
form on the subnet server, which is accessible for PCs within
the local area network (LAN) via a standard browser.
Fig
1b
click =
zoom
The main
advantage of the web form is the easy selection of all parameters,
which are important for image reconstruction (matrix size,
selecting images/frames for reconstruction, adding images/frames,
iteration steps, subsets, normalization factor, filtering,
etc.). The reconstruction program provides the page 7 following
four iterative reconstruction algorithms:
·
maximum likelihood expectation maximization (MLEM) (7)
· weighted least squares (WLS) (8)
· image space reconstruction algorithm (ISRA) (9-10)
· space alternating generalized expectation maximization
(SAGE) (11)
Each algorithm can be used together with the ordered subsets
(OS) method in order to enhance the reconstruction speed (12).
Furthermore, based on the approach of Green, the median root
prior (MRP) method as described by Alenius et al. is implemented
as an option for all reconstruction methods (13-15). Attenuation
correction can be performed either with the attenuation correction
files provided by the PET system or an iteratively reconstructed
attenuation correction map, using the MLEM algorithm with
5 iterations, 128*128 matrix, and mrp=0.3.
We have
had used phantom studies to optimize the reconstruction program.
Furthermore, phantom studies are also used on a regular basis
to check the system quality. However, several effects like
respiration movement, tissue heterogeneity and the irregular
shape of organs and structures are difficult to simulate with
phantoms, but are important for the optimization of the reconstruction
method for routine clinical use. Therefore, we selected a
standard dynamic FDG patient study to evaluate the properties
of the reconstruction program. The injected dose of FDG is
generally calculated according to the individual body weight.
Furthermore, the plasma glucose level is checked in each patient
immediately prior to the FDG application. According to our
experience, the shape of the liver FDG uptake curve does show
little variation in most of the patients, provided that diabetic
patients are excluded from the examination. However, differences
usually exist for malignant lesions due to treatment, histology,
size of the lesions, etc. Furthermore, attenuation may differ
according to the individual body shape, resulting in a large
variation of the PET image quality, which may limit the quantitative
assessment. Being aware about these parameters, we selected
a routine patient study demonstrating two adjacent, small
liver metastases. The original acquisition data were reconstructed
with all four algorithms and different reconstruction parameter
settings. The performance of the reconstruction algorithms
and the effect of different reconstruction parameters on the
high and low uptake areas were quantitatively evaluated.
A dynamic
study of a patient with two small metastases (diameter 7-8
mm according to ultrasound) in the ventral part of the right
liver lobe due to a colorectal carcinoma was selected to assess
the performance characteristics of the iterative image reconstruction.
The PET examination was performed for diagnostic purpose prior
to chemotherapy to assess the metabolic activity of the malignant
lesions already detected with ultrasound. Following positioning
of the patient, a transmission scan was performed for ten
minutes.The patient had a body weight of 70 kg and received
262 MBq FDG immediately following transmission scanning without
repositioning of the patient. The blood glucose level was
checked prior to tracer injection and was within the normal
range. The standard dynamic PET FDG acquisition protocol was
used, comprising 23 frames with 10x60 sec, 5x120 sec, and
8x300 sec. Sixty-three cross sections with an image matrix
of 256*256 pixel are reconstructed per frame. The theoretical
slice thickness is 2.425 mm per slice and comparable to the
theoretical pixel size (2.277 mm) in the cross section. The
image reconstruction settings used for the routine patient
FDG study evaluation at our center includes the reconstruction
of a summed frame, comprising the last four frames of the
dynamic series, covering the time interval from 40-60 min
post tracer application. This summed frame of the 40-60 minute
time interval is routinely used from the physicians for the
qualitative and quantitative evaluation, besides the quantitative
assessment of the whole dynamic series. We selected one cross
section from the summed frame, which demonstrates both small
liver metastases. All four iterative reconstruction algorithms
and different parameter settings for subsets, MRP, etc. were
applied to the data. Regions-of-interest (ROIs) were placed
in the cross section for one of the two page 9 metastases
(9 pixel) and for the normal liver parenchyma (425 pixel)
using a dedicated data analysis program (16-17).
Fig
2
click =
zoom
The total
number of counts was 6550055 for the slice used for the data
evaluation. Mean, standard deviation, and noise (percentage
of standard deviation) were calculated from the ROIs following
iterative reconstruction of the cross section with different
reconstruction algorithms and parameter sets.
|
