Keywords
lymphoscintigraphy - optimization - image quality - SPECT/CT - image reconstruction
Introduction
Currently, around two-thirds of cancer cases are situated in the lower abdominal and
pelvic regions. Cancer cells that metastasize often travel through the bloodstream
and tend to settle in lymph nodes and channels.[1]
[2] Understanding the extent of involvement and pathways of lymph nodes is crucial for
guiding surgical interventions. Accurate diagnosis through imaging and surgery plays
a pivotal role in effective cancer treatment.[3]
[4] With the development of new tools and techniques, imaging procedures have been accompanied
by significant progress, leading to better cancer diagnosis and prevention of additional
surgery.[5] These methods are used to detect tumors and other abnormalities to diagnose the
presence of disease and determine the effectiveness of treatment. Cancer that starts
in the cervix can spread to the lymph nodes in the pelvis.[6] The most common method for examining lymph nodes before surgery is to use a lymphoscintigraphy
scan.[7] Because of the small size of lymph nodes, the diagnosis of them in this scan is
along with the false positive or false negative.[8]
[9] This scan can utilize single-photon emission computed tomography/computed tomography
(SPECT/CT) to evaluate lymph nodes before surgery,[10] providing information on the level of involvement and the anatomical location of
these. While this method generally has a low false negative rate, errors can still
exist due to factors like attenuation, scattering, and noise, which can degrade image
quality.[11]
[12] Lymphoscintigraphy may not show any lymphatic drainage in many patients. Although
the false negative rate in the SPECT/CT technique is low, its probability is not zero.[13]
[14] Therefore, the images must have the highest quality to prevent errors and misdiagnosis.
Tissue attenuation, scattering, and noise cause the quality of images to decrease.[15]
[16]
[17] Reconstruction methods affect the quality of images. The number of iteration, subset,
and type of filter could alter the image quality.[18] Using the more subiterations and post-smoothing filter like Gaussian and Butterworth
make the images smoother with lower detectability of small objects. Also, attenuation
correction, scatter correction, and resolution recovery (RR) are often performed during
image reconstruction to achieve higher quality.
The present study aimed to find the image reconstruction protocols that reduce the
false negative or positive diagnosis in the SPECT/CT lymphoscintigraphy scan. All
image reconstruction parameters were optimized using quantitative analysis and visual
assessment.
Methods
Data Collection
The National Electrical Manufacturers Association image quality phantom containing
six fillable spheres with diameters of 10, 13, 17, 22, 28, and 37 mm was used in the
phantom study part of our research.[19] The spheres were filled with a homogenous [99mTc]Tc-pertechnetate solution at an activity concentration of 70 kBq/mL. Phantom imaging
was acquired in two conditions, first, when the background of the body was air and
the second when the background had been filled with about 7 KBq/mL of [99mTc]Tc-pertechnetate
solution to obtain sphere-to-background ratio 10:1.
Fifty female patients, between 27 and 72 years, an average of 56 years, who underwent
lymphoscintigraphy and SPECT/CT of pelvic, were selected. Seventy-five percent of
them were postmenopausal. All patients suffered from malignancies affecting the uterus,
ovaries, and cervix. Pelvic lymph node involvement was an important factor in choosing
the patients and some parameters like age, weight, and stage of disease were neglected.
The images of patients were retrospectively collected and retrieved from the hospital's
picture archiving and communication system.
Image Acquisition
Data acquisition of the phantom and patients was performed using the dual-head GE
Healthcare Discovery NM/CT 670 CZT system with low-energy high-resolution collimators.
Imaging parameters of CT images were 80 kV tube voltage, 50 mA tube current, 2.5 mm
slice thickness, and 256 × 256 matrix size. SPECT imaging was acquired in 120 projections
at 25 seconds per projection, step-and-shoot mode, and 128 × 128 matrix size.
Image Reconstruction
Vendor-provided Xeleris software was used for image reconstruction. According to Xeleris
options, all raw data of SPECT images were reconstructed using the ordered subsets
expectation-maximization (OSEM) algorithm with a combination of three iterations 2,
4, and 8, and subsets 10, 12, and 14. Gaussian filter with three different full width
at half maximum (FWHM) 2, 4, and 6 mm and Butterworth filter with 0.5, 1, and 1.5
cutoff and power of 5, 10, 15, and 20 were separately applied on the images as the
post-smoothing filters. Attenuation and scatter corrections were performed in all
SPECT images. Additionally, all SPECT images were reconstructed once more applying
the RR to assess the RR algorithm. In total, 270 SPECT images were obtained by applying
different reconstruction settings. Quantitative and qualitative assessment of phantom
images was carried out for the sphere with the 13-mm diameter because of the similarity
of the size of lymphatic nodules in clinical situation. The reconstruction settings
employed in our study are demonstrated in [Fig. 1].
Fig. 1 The flowchart of reconstruction parameters was used in this study. Reconstruction
parameters include different subiterations with various full width at half maximum
(FWHM) of Gaussian filter or using Butterworth filter with different power and cutoff.
All reconstruction sets were repeated using the resolution recovery (RR) algorithm.
Image Preprocessing
Reconstructed images should be preprocessed before quantitative analysis. We created
a three-dimensional (3D) model for the sphere's phantom segmentation using IGT module
in 3DSlicer 5.2.2. The segment editor tool of 3DSlicer 5.2.2 was utilized for lymph
node delineation in SPECT images. The volume of interest (VOI) of spheres and lymph
nodes represents the cumulative activity in the selected region. The average and maximum
count of each VOI, average count of background, and standard deviation of background
count were measured using the quantification module and segment statistic tool in
3DSlicer 5.2.2.
Quantitative and Qualitative Analysis
Three important metrics were calculated for evaluation of image quality in the sphere
with 13 mm diameter of phantom and involved lymph node in patients; contrast-to-noise
ratio (CNR), contrast, and noise (coefficient of variation [CV%]). CNR was calculated
using [Eq. (1)]:
where Cmean(sphere/nodule) is the average count in the region of interest (ROI) for the sphere
in the phantom and nodule in the patient, Cmean(background) is the average count of the spherical VOI in the background of the
body phantom or the tissue surrounding the lymph node, and SD is the standard deviation
of the background VOI. Contrast was computed with [Eq. (2)]:
where Cmax(sphere/nodule) is the maximum count in ROI, and Cmean(background) is the average count of the spherical VOI in the background of the
body phantom or the tissue surrounding the lymph node.
CV% as a statistical noise of image was calculated as follows [Eq. (3)]:
Two nuclear medicine specialists and one expert medical physicist visually evaluated
the overall image quality and detectability of the lymphatic nodule in the pelvis
and the sphere in the phantom. All images achieved from different reconstructions
were qualitatively assessed and the best one was introduced.
Statistical Analysis
All data were statistically analyzed using GraphPad Prism 10 software. CNR, contrast,
and CV% obtained from all subiterations were compared using paired t-test analysis. After selecting the subiterations that showed significantly different
from others, various parameters of the Butterworth (cutoff and power) and Gaussian
(FWHM) filters were compared using the paired t-test in selected subiterations. The significance level of the p-value was set to less than 0.05 for all comparisons. The percentage difference of
quantitative metrics was calculated to compare the role of the RR algorithm in image
quality using [Eq. (4)]:
Results
Phantom
Quantitative Analysis
It is not feasible to calculate CNR and CV% when the body phantom background is devoid
of any material, so only contrast can be calculated. Statistical analysis revealed
a significant difference between subiterations 4 × 10, 4 × 12, and 4 × 14 in the calculation
of CNR, contrast, and noise. The impact of the RR algorithm on the SPECT images was
evaluated by the percentage difference of each index under two conditions; the presence
of [99mTc]Tc-pertechnetate in the background and no radioactive material in the background
of the body phantom.
[Table 1] represents the image reconstruction sets most affected by the RR algorithm. Also,
the percentage difference of calculated metrics for these reconstruction settings
is demonstrated. RR algorithm improved CNR and contrast by 74% and more than 30%,
respectively. Additionally, image noise considerably decreased by approximately 35%
using the RR algorithm.
Table 1
Comparison of using the RR algorithm by calculating the percentage difference of image
quality metrics in two SPECT images (with and without using the RR algorithm)
|
Background of body phantom
|
index
|
Image reconstruction set
|
Percentage difference
|
|
None
|
Contrast
|
2 × 14, Butterworth (power 20, cutoff 0.5)
|
+30
|
|
[99mTc]Tc-pertechnetate
|
CNR
|
4 × 10, Butterworth (power 20, cutoff 1)
|
+74
|
|
[99mTc]Tc-pertechnetate
|
Contrast
|
2 × 14, Gaussian (FWHM 4)
|
+40
|
|
[99mTc]Tc-pertechnetate
|
CV%
|
8 × 14, Butterworth (power 20, cutoff 1.5)
|
–35
|
Abbreviations: CNR, contrast-to-noise ratio; CV%, coefficient of variation; FWHM,
full width at half maximum; RR resolution recovery; SPECT, single-photon emission
computed tomography.
Based on the results of scientific studies in this area, increasing CNR along with
reducing contrast.[8]
[20] Comparisons of both metrics in the present study as a function of the reconstruction
settings are shown in [Fig. 2]. Regardless of the type of post-smoothing filter, subiterations 40 to 56 (4 × 10,
4 × 12, and 4 × 14 in our study) can produce high-quality SPECT images. As shown in
[Fig. 2A–C], while the Butterworth filter is applied, the appropriate parameters to achieve
the best overall image quality are recommended to be 1 to 1.5 cutoff and 10 to 15
power. Also, according to [Fig. 2D], a Gaussian filter with 4 mm FWHM seems suitable based on the tradeoff between CNR
and contrast.
Fig. 2 Contrast-to-noise ratio (CNR) and contrast tradeoff plots as a function of reconstruction
sets. Dash lines represent CNR and dotted lines show contrast in all plots. (A) Represents all subiterations (i × s) with Butterworth filter using the cutoff of
0.5 and different powers, (B) represents all subiterations (i × s) with Butterworth filter using the cutoff of
1 and various powers, (C) demonstrates all subiterations (i × s) with Butterworth filter using the cutoff
of 1.5 and different powers, and (D) shows all subiterations (i × s) with full width at half maximum (FWHM) 2, 4, and
6 mm of Gaussian filter.
Qualitative Analysis
Visual assessment was performed by a medical physicist and two nuclear medicine specialists.
According to [Fig. 3A], the most appropriate reconstruction setting was subiteration 4 × 12 with the Butterworth
filter using cutoff of 0.5 or 1 and power of 5 and 10 in the presence of [99mTc]Tc-pertechnetate
in the phantom background. Also, 2 mm FWHM of Gaussian filter can be used for smoothing.
As shown in [Fig. 3B], the best reconstruction set was subiteration 4 × 12 with Butterworth filter in
the cutoff of 0.5 and 10, 15, and 20 of the power or cutoff of 1 and power of 5 in
the phantom without any radioactive material in the background. However, the Gaussian
filter with 4 mm FWHM was suitable too.
Fig. 3 Transverse slices of National Electrical Manufacturers Association (NEMA) body phantom;
(A) with filling [99mTc]Tc-pertechnetate in the background and (B) without any radioactive material in the background. The sphere with a 13-mm diameter
is mentioned with yellow arrows. The selected reconstruction settings by the expert
physicist and physicians are represented in red boxes.
Patients
Quantitative Analysis
The effect of the RR algorithm on the SPECT images was assessed by calculating the
percentage difference of CNR, contrast, and CV%. [Table 2] represents the image reconstruction settings most influenced by the RR algorithm.
RR algorithm enhanced overall image quality resulting in a 42.2% reduction of noise,
74% CNR improvement, and 37.2% contrast enhancement.
Table 2
Comparison of using the RR algorithm by calculating the percentage difference of CNR,
contrast, and, CV% in the SPECT images of patients
|
Index
|
Image reconstruction setting
|
Percentage difference
|
|
CNR
|
4 × 10, Butterworth (power 20, cutoff 1)
|
+74
|
|
Contrast
|
4 × 14, Butterworth (power 5, cutoff 1)
|
+37.2
|
|
CV%
|
8 × 14, Butterworth (power 10, cutoff 1.5)
|
–42.2
|
Abbreviations: CNR, contrast-to-noise ratio; CV%, coefficient of variation; RR resolution
recovery; SPECT, single-photon emission computed tomography.
Statistical analysis showed a significant difference between four subiterations 4 × 10,
4 × 12, 8 × 10, and 8 × 14, and the other subiterations with p-values 0.0019, 0.022, 0.069, and ≤ 0.001, respectively. No significant differences
were observed between the four mentioned subiterations.
As [Fig. 4] shows, more subiterations create more CNR values. The highest CNR was obtained in
power of 20 and cutoff of 1.5 by applying the Butterworth filter. Increasing the power
and cutoff raises the CNR value in all subiterations. The difference between the lowest
and the highest CNR in our investigated reconstruction sets was approximately 37%.
Fig. 4 Bar chart plots of contrast-to-noise ratio (CNR) in different reconstruction settings.
Comparison between four subiterations (4 × 10, 4 × 12, 8 × 10, and 8 × 14) using Butterworth
filter with different cutoffs and (A): power of 5, (B): power of 10, (C): power of 15, and (D): power of 20. Each plot is related to an individual power and each cutoff is shown
with a light to dark blue color.
According to [Fig. 5], more subiterations lead to less contrast values. The lowest contrast was obtained
in the power of 20 and cutoff of 1.5 by applying the Butterworth filter. Increasing
the power and cutoff could decrease the contrast value in all subiterations, albeit
the role of cutoff is more considerable than power. The difference between the lowest
and the highest contrast in our examined reconstruction sets was approximately 55%.
Fig. 5 Bar chart plots of contrast in different reconstruction settings. Comparison between
four subiterations (4 × 10, 4 × 12, 8 × 10, and 8 × 14) using Butterworth filter with
different cutoffs and (A): power of 5, (B): power of 10, (C): power of 15, and (D): power of 20. Each plot is related to an individual power and each cutoff is shown
with a light to dark red color.
[Fig. 6] illustrates the image noise reduced using more subiterations. The lowest CV% was
obtained in subiteration 8 × 14 by applying the Butterworth filter with the power
of 10 and cutoff of 1.5. The noise of the image could be reduced by approximately
51% if the appropriate cutoff has been selected.
Fig. 6 Bar chart plots of coefficient of variation (CV%) in different reconstruction settings.
Comparison between four subiterations (4 × 10, 4 × 12, 8 × 10, and 8 × 14) using Butterworth
filter with different cutoffs and (A): power of 5, (B): power of 10, (C): power of 15, and (D): power of 20. Each plot is related to an individual power and each cutoff is shown
with a light to dark green color.
As shown in [Fig. 7], three quantitative metrics were evaluated in four subiterations with 2, 4, and
6 mm FWHM of the Gaussian filter. Increasing subiterations does not noticeably make
a difference in the image noise, CNR, and contrast values, while FWHM of Gaussian
filter impressively affects CNR and contrast. On average, CNR increases by 65%, contrast
decreases by 36%, and noise reduces by 15% if 6 mm FWHM is used instead of 2 mm FWHM.
Fig. 7 Comparison between four subiterations (4 × 10, 4 × 12, 8 × 10, and 8 × 14) using
different full width at half maximum (FWHM) of Gaussian filter. Bar chart plots of
contrast-to-noise ratio (CNR), contrast, and coefficient of variation (CV%) in different
reconstruction settings are demonstrated in (A), (B), and (C) plots, respectively.
Qualitative Analysis
Two nuclear medicine specialists clinically assessed images of all patients obtained
from various reconstruction methods. The best image reconstruction protocol between
four selected reconstruction sets was used for qualitative assessment. The transaxial
images of one patient in different parameters of Butterworth or Gaussian filters in
subiteration 4 × 12 are demonstrated in [Fig. 8]. Preferred parameters by physicians in using the Butterworth filter tended to have
less cutoff and more power values.
Fig. 8 A transverse slice of one patient's reconstructed image using subiteration 4 × 12
with different parameters of Butterworth and Gaussian filters. The yellow arrow is
the nodule and the blue arrow is the background in the surrounding tissue near the
injection site. Preferred reconstruction sets by physicians are demonstrated in the
red box.
Discussion
Generally, reconstruction sets using the high subiteration numbers, Gaussian filter
with more FWHM and Butterworth filter with more cutoff and power, make images smoother.[18] Our findings show that CNR increased and image noise decreased by increasing the
number of subiterations using smoother filters, while contrast is reduced in this
situation. Also, overall image quality was improved using RR algorithm. RR algorithm
could increase CNR and decrease contrast and noise almost 74, 40, and 35%, respectively
([Table 1]). The effect of RR on reducing the imaging time and assessing image quality in cardiac
acquisition was evaluated by Ismail and Mansor. He showed that RR gives better quantitative
evaluation and results in a good resolution in the myocardial perfusion images.[21] Another study in 2021 demonstrated that using attenuation and scatter correction
and RR algorithm in the OSEM reconstruction method could noticeably enhance spatial
resolution in SPECT/CT of a Jaszczak phantom.[22] A retrospective study in 2022 showed RR improves spatial resolution and CNR in the
bone SPECT images.[23]
According to the inverse relationship between CNR and contrast, the higher the CNR
the lower the contrast. This results in a loss of detectability especially in small
objects. Therefore, the tradeoff between CNR and contrast in the phantom images filled
with [99mTc]Tc-pertechnetate in the background that demonstrated the most appropriate reconstruction
setting is sub-iteration 4 × 12 using Butterworth filter with cutoff of 1 and power
of 10 or Gaussian filter with 2 or 4 mm FWHM ([Fig. 2]). Based on our quantitative and qualitative analysis, it seems that the appropriate
image reconstruction setting be subiteration 4 × 12 using Butterworth filter with
cutoff of 0.5 and power of 5/10 or cutoff of 1 and power of 5/10 or 2 mm FWHM of Gaussian
filter in the phantom with radioactive material in the background. Also, the appropriate
image reconstruction setting in phantom without any radioactivity in the background
is subiteration 4 × 12 using the Butterworth filter with cutoff of 0.5 and power of
10/15/20 or cutoff of 1 and power of 5 or 4 mm FWHM of Gaussian filter. As many research
papers demonstrate, increasing the number of iterations, subsets, FWHM of Gaussian
filter, and the more cutoff and power of Butterworth filter made higher CNR but lower
contrast and noise.[24]
[25]
[26] Increasing the iteration numbers, cutoff frequency in the Butterworth filter, and
reducing FWHM of the Gaussian filter leads to enhanced spatial resolution in the SPECT
of a Jaszczak phantom.[22] Fukami et al optimized the number of iterations in SPECT images of thoracic spine
phantom. They evaluated contrast, noise, and standardized uptake value as a quantitative
analysis. Their results showed that increasing iteration numbers caused an increase
in CV% and contrast. However, contrast almost converged uniformly in subiterations
of more than 50.[27] The difference between this study and our work was the type of phantom used and
the data analysis methods.
The proposed image reconstruction parameters based on the evaluation of patient images
were subiteration 4 × 12, 4 mm FWHM of Gaussian filter, and Butterworth filter with
cutoff of 1 and power of 10/15. Our finding is in agreement with the results of a
study by Lanfranchi et al that recommended the optimized image reconstruction setting
for SPECT of brain Dat-scan. They applied the Butterworth with a 0.96 cutoff.[9] Another study by Lyra and Ploussi on SPECT of the liver showed that the Butterworth
filter with the cutoff of 0.1 to 0.5 could help to achieve the high detectability
of small lesions.[28] We found the cutoff of 0.5 in the Butterworth filter as the optimized parameter
in the reconstruction protocol, similar to the Lyra and Ploussi research. Two studies
utilized the SPECT/CT images of cardiac[10] and bone scan.[14] In the myocardial perfusion scan, Gaussian filter with 12 to 14 mm FWHM was applied
to smooth the images.[10] Optimized filters for the assessment of bony metastatic lesions were proposed 10
to 13 mm FWHM of Gaussian filter by Alqahtani et al.[14] These results were in contrast with our findings because of the object sizes. Extra
smoothing of the images could lead to misdiagnosis and false negatives when the lesion
size is small, like lymph nodes. While in large lesions such as obvious bone metastasis
or cardiac muscle, more smoothing makes CNR better, reduces noise, and improves image
quality without any false negativity. According to our results, 2 and 4 mm FWHM of
Gaussian filter were recommended in lymphoscintigraphy SPECT/CT of pelvic. Choosing
the appropriate reconstruction parameters strongly depends on the size of the lesions.
Overall image quality does not degrade using the smoother filter in evaluating large
objects, while it is influenced on detectability and assessment of small lesions.
Although many studies about the optimization of image reconstruction settings have
been performed, none of them simultaneously assessed all parameters. We evaluated
iterations, subsets, two different smoothing filters, and the RR algorithm to find
the optimized reconstruction sets for small lesions. Nonetheless, our study had some
limitations. First, we studied the optimization of reconstruction sets only for patients
with lymph nodes involved in pelvic and these results cannot be generalized to other
cases with different sizes of lesions. Second, we optimized the reconstruction parameter
options that exist in one make and model of gamma camera scanner (GE Discovery 670
CZT), while it can be evaluated for other gamma cameras as a multicenter study similar
to harmonization in positron emission tomography/CT scanner.[29]
[30]
Conclusion
According to our findings, subiteration 4 × 12 proposed with the Butterworth filter
using a power of 10 and a cutoff of 1 or 4 mm FWHM of the Gaussian filter. Furthermore,
utilizing the RR algorithm in the Discovery NM/CT 670 GE gamma camera is recommended
to achieve the highest image quality in lymphoscintigraphy scans.
