Keywords
uniformity - ImageJ - NMQC Toolkit - NM Toolkit - SPECT
Introduction
Nuclear medicine imaging retrieves details regarding the physiological function and
radiopharmaceutical distribution within a specific organ. The reliability of NaI (Tl)
based single photon emission computed tomography (SPECT) detectors is influenced by
several factors, including potential errors arising from the conversion of gamma photons
into electrical signals, leading to issues like photon loss, motion artifacts due
to extended acquisition time, and increased radiation exposure.[1] Common sources of artifacts in SPECT imaging encompass nonuniformity in gamma camera
detectors, center of rotation errors, misalignment of cameras in multidetector scanner
systems, reconstruction error, patient movement, unintended uptake of radiotracers
in other organs, and attenuation.[2] Among these factors, uniformity emerges as the most crucial gauge of SPECT performance,
prompting daily performance evaluations. Even a minor flaw in detector uniformity,
as minimal as a 3% disparity, can result in a noticeable artifact in the reconstructed
image.[3] Notably, most issues linked to the integrity of the detector head, computer system,
and hard copy device can be identified through the uniformity image. Gamma camera
system nonuniformity typically manifests as conspicuous concentric ring artifacts,[4]
[5] often necessitating recalibration of the detectors.[5]
In SPECT imaging, numerous physical parameters influence image uniformity, necessitating
the consideration of each parameter before commencing imaging. Nonuniformities may
stem from diverse factors, encompassing fluctuations in the pulse-height spectrum
of the photo multiplier tubes (PMTs),[6] spatial nonlinearities,[7] and other issues such as incorrect energy settings, flawed linearity maps, poor
PMT balance, and scintillation crystal hydration.[8] Consequently, achieving standardization and automation in SPECT proves challenging.
Determining the threshold for gamma camera nonuniformity poses difficulty due to unknown
reproducibility in quality control (QC) measurements and service adjustments. In nuclear
medicine, sustaining optimal gamma camera performance typically demands stringent
adherence to QC tests, often aligning with international standards sets by entities
like the National Electrical Manufacturers Association (NEMA), American Association
of Physicists in Medicine (AAPM), Society of Nuclear Medicine and Molecular Imaging
(SNMMI), and European Association of Nuclear Medicine (EANM) protocols, widely recognized
as reference guidelines.[9] Yet, implementing these protocols routinely face challenges due to their intricacies.
To alleviate the burdens associated with QC testing, manufacturers have introduced
gamma camera systems equipped with integrated software for analyzing QC outcomes,
facilitating immediate assessment of the instrument's performance parameters. Most
vendor-specified or integrated QC analysis software systems are developed based on
standardized QC protocols but optimized to suit the technological specifics of the
systems, ensuring optimal functionality and results. For instance, certain gamma camera
manufacturers integrate uniformity correction, alongside energy and linearity corrections,
to rectify residual nonuniformity and collimator imperfections.[8] Furthermore, result analysis is often automated, presenting finalized outcomes without
extensive elaboration, albeit lacking user configurability.[10] The complex computation applied to the data prior to perception can introduce additional
uncertainties in various ways.[11]
[12] Previous studies have reported deviations in the agreement of built-in QC analysis
software with standard QC protocols.[13]
[14]
[15]
[16] However, the rigorous assessment of the software's agreement remains lacking. Considering
these aspects, it becomes imperative to evaluate the alignment between vendor-developed
software and NEMA standards using independent analysis QC software.
Hence, to authenticate and ensure the reliability of the integrated QC analysis software
developed by the manufacturer, this study aims to compare the intrinsic uniformity
QC outcomes generated by our institution's integrated QC analysis software with the
data derived from three distinct QC analysis software—both free and open/closed source—to
ascertain their level of concordance. This study employed three separate QC software
tools, previously cited in the literature,[13]
[17]
[18] in addition to the integrated QC software. These software packages encompass the
NMQC Toolkit, NM Toolkit, and Image J (Fiji). Among these, Image J has been widely
adopted in nuclear medicine image analysis and applications, likely owing to its open
source nature.[19]
[20]
[21]
[22]
[23]
[24]
Materials and Methods
The entire data acquisition and analysis were carried out using an integrated SPECT/CT
system (Discovery NM/CT 670 Pro, GE Medical System, United States) and its common
user interface for the uniformity analysis. Three distinct, open/closed source of
QC analysis software packages were employed to analyze the uniformity: IAEA-NMQC Toolkit
version 1.0.14,[17] Fiji – ImageJ,[25] and NM Toolkit version 1.5.24.[26]
Intrinsic Uniformity Test
In this investigation, we strictly adhered to the guidelines outlined in the NEMA
Standards Publication NU 1-2018 when evaluating the intrinsic uniformity of the gamma
camera system.[27] The collimators initially affixed to the gamma camera's detector heads were exchanged
with decoy collimators. To prepare a 99mTc point source, we utilized a tuberculin syringe containing 800 μCi of 99mTc,[13] and positioned it at a location with a distance equivalent to five times the largest
dimension of the useful field of view (UFOV) of the detector.[27] This positioning was precisely aligned with the central axis of the detector ([Fig. 1]). During data acquisition, the process was programmed to halt at 20,000 kilocounts,
employing an image matrix size of 128 × 128. The choice of the matrix size aligned
with the NEMA guideline, stipulating that the flood field image should be stored in
a matrix size generating pixel dimensions of 6.4 mm ± 30%. The energy window setting
recommended by the manufacturer for 99mTc was employed, configured at 20%, symmetrically around the photopeak. In this study,
the imaging acquisition was repeated for both detectors. After obtaining the uniformity
images, the analysis was conducted using the integrated QC software within the SPECT
system.
Fig. 1 The intrinsic uniformity setup. 99mTc point source was positioned in a holder (pointed by the red arrow) at a distance of ≥5 times the largest dimension of the uniform field of view (UFOV)
of the detector, adhering to the NEMA guidelines.
Uniformity Analysis Using Free Open/Closed Source Software
Three distinct, free and open/closed source software applications—NMQC Toolkit, Fiji
(ImageJ), and NM Toolkit—were employed. The first two software tools are available
for download from their respective website: https://humanhealth.iaea.org and https://imagej.net/software/fiji/. Meanwhile, the NM Toolkit is a closed source software provided for free, necessitating
a request from its developer.[26] It is worth noting that the analyses performed by the NMQC Toolkit and NM Toolkit
software align with the recommendations outlined in the NEMA Standards Publication
NU 1-2018. [Table 1] provides an overview of the software utilized in this research.
Table 1
The free, independent, open/closed QC analysis software tools used in this study
|
Software
|
|
NMQC Toolkit
|
Fiji (Image J)
|
NM Toolkit
|
|
Accessibility
|
Free
|
Free
|
Free
|
|
Software type
|
Closed source
|
Open source
|
Closed source
|
|
Operating system compatibility/requirements
|
Windows, Mac OS X, Intel 32-bit or 64-bit with bundled Java
|
Windows, Mac OS X, and Linux in both 32-bit and 64-bit modes
|
Windows
|
|
DICOM image format support
|
Yes
|
Yes
|
Yes
|
|
Compliance with NEMA protocols
|
Yes
|
–
|
Yes
|
|
Available tests/tools for quality control
|
Yes
|
No[a]
|
Yes
|
Abbreviations: DICOM, Digital Imaging and Communications in Medicine; NEMA, National
Electrical Manufacturers Association; QC, quality control.
a It is not specifically designed for quality control in medical imaging; it can be
used for basic image analysis and measurement tasks.
NMQC Toolkit
In NMQC Toolkit image uniformity analysis, the tool performed multiple tasks to ensure
adherence to the NEMA requirements. First, it inspected the pixel size to confirm
its alignment within the NEMA-specified range of 6.4 mm ± 30% (ranging from 4.48 to
8.32 mm). Second, if the pixel size exceeded this range, an automatic matrix size
rescaling procedure (using the sum method) would be initiated. In instances where
the pixel size was smaller than NEMA's recommended minimum value, the adjacent detector
pixels were combined to yield an effective pixel size within the specified range.
However, it is crucial to note that acquiring uniformity images with a pixel size
significantly exceeding the maximum value recommended by NEMA is discouraged. In such
cases, a warning would be generated, and pixel size reduction would not be executed.
Fiji (ImageJ)
Fiji was not specifically developed for dedicated QC and testing purposes in nuclear
medicine modalities. Nevertheless, it has gained widespread acceptance in the field
of image analysis and various applications. It is important to note that, in this
study, Fiji analysis adhered to the guidelines outlined in the NEMA standards. Fiji
was solely employed for extracting pixel values from the uniformity image. Before
pixel value extraction, the uniformity image underwent filtering using the convolution
kernel, described in [Equation 1].[27] Following this, the uniformity analysis was carried out through calculations performed
in Microsoft Excel.
During the calculation, only non-zero-pixel values were taken into account, excluding
zero-pixel values at the periphery of the image. Using Microsoft Excel, integral,
x-differential, and y-differential uniformities were computed based on the equations
provided in NEMA Standards Publication NU-1 2018. Mathematically, the two equations
seem similar ([Eq. 2]). However, the integral uniformity was determined by calculating the difference
between the maximum and the minimum pixel values within the respective fields of view
(FOVs). In contrast, the differential uniformity involved finding the difference between
the maximum and minimum pixel values within a set of five contiguous pixels in a row
or column.[27]
NM Toolkit
The NM Toolkit facilitates the automated assessment of uniformity analysis. To execute
the analysis, the Digital Imaging and Communications in Medicine (DICOM) image was
uploaded in the “Image Folder” and chosen for analysis. Subsequently, by simply clicking
on the “NEMA Uniformity” option, the software displayed the postprocessed intrinsic
uniformity image along with the test results.
Statistical Analyses
This study used nonparametric Mann–Whitney U test to ascertain whether the differences between the integral and differential uniformities,
obtained using integrated QC analysis software and three distinct types of free QC
analysis software were statistically significant. To assess the agreement between
the software, a Bland–Altman regression approach was employed as the computed data
did not exhibit a normal distribution.[28]
[29]
Results
The intrinsic uniformity computed by integrated QC software of Discovery NM/CT 670
Pro scanner and the other three free open/closed software packages (NMQC Toolkit,
Fiji, and NM Toolkit) for both central field of view (CFOV) and UFOV are illustrated
in [Fig. 2]. Percentage deviation was estimated in terms of percentage difference with respect
to the uniformity values computed by the SPECT integrated QC software. Minor deviations
between the integrated QC software and free open/closed software were observed, less
than 3% for the integral uniformities and less than 2% for differential uniformities
(see [Table 2]).
Fig. 2 Comparison of the uniformities computed using integrated QC software and free QC
software. (a, b) Integral uniformity. (c, d) X-differential uniformity. (e, f) Y-differential uniformity. Left panels: central field of view. Right panels: uniform
field of view.
Table 2
Comparison of the uniformity, presented as a percentage deviation from the SPECT integrated
QC software (standard deviation)
|
|
NMQC Toolkit
|
Fiji
|
NM Toolkit
|
|
Integral uniformity (%)
|
CFOV
|
2.46 (0.18)
|
2.56 (0.13)
|
2.44 (0.19)
|
|
UFOV
|
2.34 (0.19)
|
2.64 (0.30)
|
2.31 (0.18)
|
|
x-differential (%)
|
CFOV
|
1.91 (0.21)
|
1.96 (0.19)
|
1.91 (0.21)
|
|
UFOV
|
1.84 (0.22)
|
1.90 (0.21)
|
1.83 (0.23)
|
|
y-differential (%)
|
CFOV
|
1.77 (0.24)
|
1.79 (0.18)
|
1.78 (0.24)
|
|
UFOV
|
1.69 (0.26)
|
1.93 (0.50)
|
1.69 (0.27)
|
Abbreviations: CFOV, central field of view; QC, quality control; SPECT, single photon
emission computed tomography; UFOV, useful field of view.
Agreement between the Estimated Uniformity
The significant difference between the intrinsic uniformity computed by integrated
QC software and each free software was confirmed by the Mann–Whitney U test (z = −3.782 to −3.780, p < 0.001; [Table 3]). The other results were purposely not presented here due to the minor difference
with the data presented here. In this study, the free software consistently resulted
in higher uniformity compared to integrated QC software, presented by the higher mean
rank (mean rank free software = 15.5, mean rank integrated QC software = 5.5).
Table 3
Mann–Whitney U test for uniformity analysis (integral and differential uniformity)
|
Variables
|
Group (N)
|
Mean rank
|
Z
|
Sig.
|
|
Uniformity (%)
|
Integrated QC software (10)
|
5.5
|
−3.782 to −3.780
|
<0.001
|
|
Free open/closed software (10)
|
15.5
|
|
|
Note: The same outcomes were obtained for all free open/closed software compared to
integrated QC software.
[Figs. 3]
[4]
[5] present the regression-based limits of agreement for intrinsic uniformity in both
CFOV and UFOV between the integrated QC software and free open/closed software. A
good level of agreement was observed between the data. Most of the measurements were
observed to be within the 95% limits of agreement, which were plotted as the upper
and lower limits (dotted line). Observation of the slopes of the regression lines demonstrated that they are not
exactly 45 degrees, indicating one method measured proportionately more or less than
the other method. In this study, the free open/closed source QC analysis software
resulted in higher percentages of intrinsic uniformity than the values computed by
the integrated QC software in Discovery NM/CT 670 Pro.
Fig. 3 Bland–Altman regression-based limits of agreement for integral uniformity computed
between (a, b) integrated quality control (QC) software and NMQC Toolkit; (c, d) integrated QC software and NM Toolkit; (e–f) integrated QC software and Fiji. A good level of agreement was observed, with the
data falling within the 95% limits of agreement. The solid line indicates the mean difference (bias), and the dotted lines represent 95% limit of agreement. Left panels: central field of view. Right panels:
uniform field of view.
Fig. 4 Bland–Altman regression-based limits of agreement for x-differential uniformity computed
between (a, b) integrated quality control (QC) software and IAEA-NMQC Toolkit; (b, c) integrated QC software and NM Toolkit; (e, f) integrated QC software and Fiji. A good level of agreement was observed, with the
data falling within the 95% limits of agreement. The solid line indicates the mean difference (bias), and the dotted lines represent 95% limit of agreement. Left: central field of view. Right panels: uniform
field of view.
Fig. 5 Bland–Altman regression-based limits of agreement for y-differential uniformity computed
between (a, b) integrated quality control (QC) software and IAEA-NMQC Toolkit; (b, c) integrated QC software and NM Toolkit; (e, f) integrated QC software and Fiji. Left panels: A good level of agreement was observed,
with the data falling within the 95% limits of agreement. The solid line indicates the mean difference (bias), and the dotted lines represent 95% limit of agreement. Left panels: central field of view. Right: uniform
field of view.
Discussion
SPECT, in comparison to planar imaging, offers superior localization of the radioactivity
distribution. However, it faces challenges related to nonuniformity stemming from
the back projection of data during image reconstruction.[30] Therefore, to ensure the optimal functioning of a SPECT system, various QC procedures,
coupled with correction techniques, are implemented to address and enhance its performance.
One of the critical factors to consider is the uniformity of the system itself, measuring
its ability to generate a uniform image when exposed to uniform gamma rays. While
some accept smaller nonuniformity within the 1 to 3% range,[31] our SPECT/CT allows for a slightly larger value, less than 5%. However, even a minor
flaw in detector uniformity, as minimal as a 3% disparity, can lead to noticeable
artifacts in the reconstructed image.[3] In line with this, some study suggests considering every component of the SPECT
or gamma camera system for proper uniformity correction, including external factors
such as the contribution from the collimator.[32]
In this study, acquisition of uniformity image strictly adhered to the NEMA recommendations
with the goal of assessing variations resulting solely from different analysis tools.
The standard protocol for the uniformity assessment typically follows the NEMA guidelines,
which provide standard recommendations. According to the NEMA guidelines, the assessment
should use the energy window recommended by the manufacturer for the selected acquisition
protocol, as the energy window significantly influences flood image uniformity. In
99mTc imaging, 20% photopeak window is commonly used, as studies have shown that a range
of 15 to 20% yields optimal uniformity. Deviating from this range, either with a lower
or higher photopeak window, has been demonstrated to associate with marked changes
in uniformity.[2] In addition to that, to synchronize readings from two detectors, strict adherence
to NEMA recommendations is crucial, ensuring a minimum of 10,000 counts collected
in the center pixel of the image.
Uniformity assessment in this study resulted in nonuniformity of less than 5%, irrespective
of the analysis tools used, thus complying with recommendations.[30] However, the difference between the data and the factors associated with it should
be understood. Discrepancies were noted with respect to the QC software integrated
into the scanner system, with greater nonuniformity estimated by the free open/closed
software ([Table 2] and [Fig. 2]), aligning with previous findings.[13] This significant difference could be attributed by two factors. First, it is attributed
to the correction methods applied by the SPECT system on the acquired image. Some
gamma camera manufacturers, as per the IAEA, use uniformity correction, in addition
to energy and linearity corrections, to correct for residual nonuniformity and collimator
flaws.[8] In this study, the intrinsic uniformity computed by the integrated QC software was
based on images that underwent postprocessing with uniformity correction by its system.
Previously, a study highlighted the difference in intensity profiles of a flood-field
image before and after uniformity correction, emphasizing the improvement in the standard
deviation of the profile after correction.[6] However, extracting these postprocessed images from our system was not possible.
For the analysis using independent software, a raw intrinsic uniformity image was
utilized, and these images were normalized through convolution with the NEMA-defined
kernel.[27] Hence, the difference between the integrated QC software and other free software
was expected. Second, it could be attributed to the pixel size used during the analysis.[13]
[33] Some analysis tools like NM toolkit used default pixel size, for which alteration
is not possible. Even though the NEMA allows pixel size in the range of 4.48 to 8.32 mm,
the wide range of this value is believed to affect the pixels counts due to the smoothing
effects, subsequently influencing the uniformity estimation.[13]
A comparison of the free open/closed software shows that Fiji resulted in larger deviation
from the integrated QC software (albeit <3% deviation and the majority of the computed
values are overlapping each other), probably due to manual analysis done in Fiji.
Fiji involved importing pixel values to Excel for computation (with reference to the
NEMA guidelines), while others were automated computations, which display the finalized
results without much, if any, elaboration on them. Therefore, Fiji analysis could
be subjected to random and systematic errors due to the inherent variability in any
sampling process during pixel selection process. Although the estimated nonuniformity
was less than 5% and compliant with recommendations,[30] understanding the differences between the data and associated factors is crucial.
The Bland–Altman regression-based limit found in this study indicates good agreement
between the integrated QC software and free open/closed source software in intrinsic
uniformity analysis. Most measurements fell within the 95% limit of agreement, although
there was at most one outlier for some plots ([Fig. 3a] and [c]). The outliers observed were in integral uniformity, as its assessment was performed
for the entire flood to assess the overall performance of the system. However, integral
uniformity lacks detailed spatial information and may not be sensitive to localized
or small-scale variations. Pixel averaging performed on the image can mask variations
at the pixel level and might not be sensitive to small changes in intensity. When
observing the slopes of the regression lines, they are not exactly 45 degrees, indicating
one method measured proportionately more or less than the other method.[34] In this study, the free open/closed QC analysis software resulted in higher percentages
of intrinsic uniformity than the integrated QC analysis software developed by the
vendor of Discovery NM/CT 670 Pro QC analysis software.
Conclusion
Manufacturers have responded to the challenges in routine QC testing by equipping
recent gamma camera systems with built-in QC analysis software, which has proven to
be highly efficient. This study aimed to assess the agreement between uniformity analysis
computed by the integrated QC software developed by manufacturers and the analysis
performed by independent QC software. The data presented in this study are, however,
limited to a single SPECT/CT scanner, the GE Discovery NM/CT 670 Pro system. Based
on regression-based limits of agreement analyses, it is concluded that the intrinsic
uniformity computed by the integrated QC software exhibits a good level of agreement
with the data computed by the NMQC Toolkit, NM Toolkit, and Fiji. It is also affirmed
that the QC analysis processes of the integrated QC software align well with the standard
QC analysis protocols recommended by the NEMA. In conclusion, the uniformity analysis
performed by the integrated QC software is reliable and yields values closely aligned
with the NEMA standards. The comparison serves as a benchmark for the performance
of the integrated QC software in assessing uniformity. The findings aid practitioners
and researchers in making informed decisions about the selection of analysis tools
for uniformity assessments. Further analysis incorporating different models of SPECT/CT
scanners would be beneficial to validate the findings across a broader range of systems
and enhance the generalizability of the results. By contributing to the ongoing efforts
in standardization within the field, this work helps establish best practices for
uniformity assessment, harmonizes procedures across different imaging centers, promotes
consistency, and facilitates multicenter studies and comparisons.
