NetForum uses cookies to ensure that we give you the best experience on our website. If you continue to use the site, we'll assume that you are happy to receive these cookies on the NetForum website. Read about our cookies.
NetForum Community
Learn. Share. Optimize.
Log in | Sign up now | Submit content | Contact
Go to similar content

Improved clinical performance of myocardial perfusion SPECT imaging using Astonish iterative reconstruction

White Paper
Philips NM Marketing Philips Healthcare

G.V. Heller - Hartford Hospital, Nuclear Cardiology Laboratory, Hartford, CT, USA.

T.M. Bateman - University of Missouri-Kansas City, Mid America Heart Institute, Kansas City, MO, USA. Cardiovascular Imaging Technologies, Kansas City, MO, USA.

S.J. Cullom - Cardiovascular Imaging Technologies, Kansas City, MO, USA.

H.H. Hines, A.J. Da Silva - Philips Healthcare, San Jose, CA, USA.

Myocardial perfusion SPECT imaging is a useful tool to help diagnose patients suspected of having coronary artery disease or to assess patients with known disease. There has been significant growth in the number of these procedures, with about 8.9 million performed in 2007. The laboratories performing these procedures are under pressure to reduce costs, improve image acquisition efficiency, reduce absorbed radiation dose and improve diagnostic accuracy. Additionally, the recent shortages of 99mTc have further increased the call for more efficient use of 99mTc-labeled radiopharmaceuticals.

To address these needs, Philips has developed an improved reconstruction method termed "Astonish". Astonish improves reconstruction accuracy and makes more efficient use of acquired counts by incorporating internal modeling of the imaging physics in the reconstruction process. Astonish also provides for scatter and attenuation correction, which can mitigate artifacts in the imaging process that can reduce lesion contrast or mimic perfusion defects.

This paper provides a brief explanation of the reconstruction methods and presents clinical results of a recent multi-center trial designed to test performance of the Astonish algorithm for myocardial perfusion SPECT imaging. The clinical performance of Astonish was tested on both full-count and half-count stress-rest data, both with and without attenuation and scatter corrections1. Clinical performance of a stress-only imaging approach that can further improve laboratory efficiency was also tested on full-count and half-count data with scatter and attenuation corrections2.

The half-count data gives the nuclear physician the option of improving efficiency by reducing the acquisition time using conventional dosing protocols, or reducing the radiation dose and, potentially, radiopharmaceutical costs by injecting less radioactivity while using more conventional acquisition times. The improved diagnostic accuracy of attenuation-corrected data enabled stress-only interpretations. Elimination of the resting portion of the exam further reduces radiation exposures and improves laboratory efficiency.

Reconstruction methods
Iterative reconstruction methods are well suited to improve image quality and clinical performance. This is done by incorporating internal modeling of the imaging physics in SPECT reconstructions to correct for the major factors affecting image quality. These factors include degradation of spatial resolution with increasing distance of the source from the collimator, Compton scatter and photon attenuation.

The degradation of spatial resolution with distance from the collimator reduces the overall spatial resolution in the images and can produce non-uniformity artifacts in the heart. Scattered photons detected in the energy window reduce image contrast and interfere with the ability to efficiently perform attenuation correction. Non-uniform photon attenuation may produce artifacts that can mimic perfusion defects if not corrected. Attenuation correction also plays an important role in differentiating artifacts from true defects with stress-only imaging. Astonish is an OSEM (Ordered Subsets Expectation Maximization) iterative method that incorporates statistical noise reduction3 and depth-dependent resolution recovery through internal modeling of the imaging physics. When an attenuation map is present, Astonish can additionally perform Compton scatter correction and attenuation correction.

Figure 1 provides a diagram of the OSEM iterative process using three ordered subsets as an example. In the iterative process, a new estimated image is calculated by multiplying the previous image by a modifier matrix to produce the subsequent image. In other words, the next image (n + 1) is produced by multiplying the previous image (n) by a modifier matrix (n). The modifier matrix is developed using subsets of the projection data as described below. The "ordered subsets" are a limited number of the total projections. For example, two ordered subsets of 64 total projections are the 32 odd projections (1, 3, 5 .63) and the 32 even projections (2, 4, 6 . 64).
Figure 1 Depiction of the OSEM iterative reconstruction process showing the relationship
between the image estimates at each step and the modifier matrix. As the image estimate
becomes “closer” to the image of the true object, the amount of change in the modifier matrix
becomes smaller.
Figure 1
Depiction of the OSEM iterative reconstruction process showing the relationship between the image estimates at each step and the modifier matrix. As the image estimate becomes “closer” to the image of the true object, the amount of change in the modifier matrix becomes smaller.

 

A key to understanding iterative reconstruction methods is the understanding of how the modifier matrix is calculated and applied throughout the process. Figure 2 depicts conceptually how the modifier matrix (shown in green) is produced in a three-step process. Note that the process is shown only for one projection angle, so these steps must be repeated for all of the angles in a subset to produce the modifier matrix for that subset.

Figure 2 The modifier matrix is calculated in a three-step
process. The first step is the forward projection (from
reconstruction plane to projection plane) of an image
consisting of pixels all having the same count values in
the image matrix to produce an estimated projection
(which is a constant value for the initial image estimate).
The next step is to compare the measured and
estimated projection values including smoothing (filtering)
of the result using the matched filter. The thir
Figure 2
The modifier matrix is calculated in a three-step process. The first step is the forward projection (from reconstruction plane to projection plane) of an image consisting of pixels all having the same count values in the image matrix to produce an estimated projection (which is a constant value for the initial image estimate). The next step is to compare the measured and estimated projection values including smoothing (filtering) of the result using the matched filter. The thir

In the first step, the values in the image matrix (shown in blue) are forward-projected to produce an estimated projection, EØ. As shown, Astonish uses a constant value as the initial image estimate.

In the second step, the estimated projection is compared with the measured projection, PØ, to produce an error estimate by taking the ratio of PØ/EØ. In this step to minimize the image noise, Astonish uses a matched filtering procedure. In the matched filtering step, both the estimated projection and the measured projection are filtered by smoothing the quantity PØ/EØ. Performing the smoothing within the iterative process preserves spatial resolution and improves uniformity of the myocardium (by minimizing the impact of image noise). By comparison, smoothing after completing the iterative process would reduce spatial resolution in the final resulting images.

In the third step, the smoothed error estimate, PØ/EØ, is back-projected into the modifier matrix. This is repeated until all the angles in a single subset are utilized. Once all of the subsets have been utilized, a single iteration has been completed. Note that while Figure 2 shows only a single representative projection angle, the process is repeated for each of the angles in a subset. This means that the modifier matrix is updated for all of the projection angles in a single subset. A single iteration is then completed when all of the subsets have been used to update the modifier matrix and all of the measured projection angles are used.

The spatial resolution of the collimator degrades with increasing distance from the surface of the collimator. Compensating for this effect can improve image quality. The Astonish algorithm performs depth-dependent resolution recovery as shown in Figure 3. The change of spatial resolution with distance from the collimator is calibrated in the software. During the acquisition of patient data, the distance from the collimator to the center of rotation is recorded for each projection angle. During the reconstruction process, the changes in spatial resolution with distance are calculated in both the forward projection in the image matrix and in the back-projection in the modifier matrix as shown in Figure 3. The counts are spread over multiple pixels during both the forward- and backprojection steps and the degree of broadening is determined by the distance between the pixel and collimator using the collimator resolution response function.
Figure 3 Resolution compensation is performed in both the back-projection in the modifier
matrix and in the forward-projection in the image matrix. The previously measured collimator
response function (shown on the right) is used to calculate the amount of broadening.
Figure 3
Resolution compensation is performed in both the back-projection in the modifier matrix and in the forward-projection in the image matrix. The previously measured collimator response function (shown on the right) is used to calculate the amount of broadening.

Non-uniform attenuation can produce attenuation artifacts in the perfusion images that mimic perfusion defects. These defects may be reduced by performing both scatter and attenuation correction. When an attenuation map is present, the Astonish algorithm can perform scatter and attenuation correction in addition to resolution recovery.

Scatter correction is performed using the ESSE method (Effective Source Scatter Estimation) [4]. This method uses the concept of defining an effective source distribution of scatter, calculating the expected scatter projection data from this effective source of scatter photons, and adding this calculated scatter projection data to the projection of the estimated image to produce the total estimated projection EØ. The estimate of the scatter is calculated from the image estimate and density map by convolving the image estimate with a three-dimensional scatter convolution kernel. This convolution kernel is pre-calculated and is dependent upon the photon energy.

The attenuation correction method is illustrated in Figure 4. Attenuation is modeled only during the forward-projection process using both the image matrix and the attenuation map to calculate the estimated projection, EØ. A tissue transmission density image is acquired with a radionuclide scanning line source or a CT scanner, either simultaneously or sequentially with the emission scan. Since the emission and transmission images typically involve different photon energies, the transmission density images are scaled to the appropriate photon energy using the mass attenuation coefficient to produce a matrix of linear attenuation values known as the "attenuation map". The attenuation map is used to calculate the amount of attenuation from each point in the object through the body in the image matrix. Figure 4 shows the path along which the counts in the image matrix are attenuated. This calculation is performed during forward projection to produce attenuated estimated projections, EØ, at each projection angle.
Figure 4 Attenuation correction is performed during the forward projection process in the
image matrix using the transmission map to calculate the amount of attenuation along the path
length through the body as indicated by the arrows.
Figure 4
Attenuation correction is performed during the forward projection process in the image matrix using the transmission map to calculate the amount of attenuation along the path length through the body as indicated by the arrows.

Clinical testing methods
The clinical imaging performance of Astonish, and Astonish with scatter and attenuation correction, was evaluated recently in a multicenter trial. Image quality, diagnostic confidence and diagnostic accuracy for detection of coronary artery disease were evaluated on data from 187 consecutive patients undergoing clinically indicated myocardial perfusion SPECT studies. The patients underwent subsequent cardiac catheterization (132 patients) or had a low likelihood of coronary artery disease (55 patients).

All studies were acquired on Philips CardioMD small field of view gamma cameras with VantageT Gadolinium-153 scanning line sources. The line sources were used to acquire the transmission data. The data were acquired according to ASNC imaging guidelines5: LEHR collimator, 64 projections at 20 seconds/projection for the stress and 30 seconds for the rest images. An 180° RAO-LPO orbit was acquired beginning at 15 - 45 minutes after injection of 30-35 mCi of a 99mTc agent for the stress studies and 10 mCi for the resting studies. Tc-99m-sestamibi was used in 176 patients and Tc-99m-tetrofosmin in 14 patients. The energy windows were set at 140 keV +/- 10% for the emission and 100 keV +/- 10% for the Gd-153 line source transmission data and 118 keV, +/- 6% for down scatter correction. The half-count study was derived from the full-count study by stripping out "every other" projection from the original 64 projection data set. The processed images were interpreted in a blinded fashion without knowledge of the number of projections used in the reconstructions and without knowing if attenuation correction was or was not applied.

The images were reconstructed using three different methods: conventional filtered back-projection, Astonish, and Astonish with attenuation correction.

Conventional filtered back-projection (FBP) images were reconstructed using a Butterworth filter for pre-filtering of the projections (order 5 and cutoff frequency of 0.45) for the perfusion images, and ECG-gated images were reconstructed in the same way, except that a cutoff frequency of 0.35 was used.

Astonish and Astonish with attenuation correction used four iterations, eight subsets and a Hanning match filter parameter of 1.0. ECG-gated images used the same number of subsets and iterations, with a Hanning match filter parameter of 0.8.

Coronary angiography
The percentage of minimal narrowing was determined visually at each lab and reported. The angiographic reports from each center were submitted to a single laboratory where the location and percentage of luminal stenosis narrowing was extracted from the reports for the left main and the three major coronary arteries and their major branches. The percentage of luminal narrowing was determined visually. Significant disease was defined at a 70% stenosis threshold in one or more major coronary arteries, or 50% stenosis in the left main artery, and this value was used for CAD diagnosis unless otherwise specified.

Results
There was a statistically significant improvement in image quality for both full-count and half-count images reconstructed with Astonish in comparison with those reconstructed with filtered back-projection (FBP) (Figure 5).

There was no statistical difference in the interpretive certainty of full-count and half-count images reconstructed with Astonish, p = 0.18. This demonstrates the ability of Astonish to achieve good image quality with half the counts (Figure 6).

There was no statistical difference in the diagnostic accuracy of full-time data reconstructed with filtered back-projection or full- or half-count imaging using Astonish (Figure 7).

There was a statistically significant improvement in specificity and normalcy when myocardial perfusion data was corrected for scatter and attenuation. This was true for both full-count and half-count data (Figure 8).
Figure 5 Comparison of the image quality of stress
perfusion data reconstructed with filtered back-projection
(full-count data) and Astonish on full- and
half-count data.Figure 6 Comparison of the interpretative certainty of half-count data reconstructed using
Astonish without and with attenuation correction.Figure 7 Comparison of the diagnostic accuracy of
images reconstructed using filtered back-projection
(FBP) and Astonish with full- and half-count data.
Figure 5
Figure 6
Figure 7
Comparison of the image quality of stress perfusion data reconstructed with filtered back-projection (full-count data) and Astonish on full- and half-count data.
Comparison of the interpretative certainty of half-count data reconstructed using Astonish without and with attenuation correction.
Comparison of the diagnostic accuracy of images reconstructed using filtered back-projection (FBP) and Astonish with full- and half-count data.
Figure 8 Comparison of the diagnostic accuracy of full- and half-count data reconstructed with Astonish and Astonish with scatter and attenuation correction.
Figure 8
Comparison of the diagnostic accuracy of full- and half-count data reconstructed with Astonish and Astonish with scatter and attenuation correction.

In the stress-only portion of the trial, patient studies were interpreted without using the rest images of the studies. When the patient studies were interpreted using the stress images reconstructed using Astonish with attenuation correction, a high degree of diagnostic accuracy was obtained as shown in Figure 9.

The stress-only data also provided a high level of diagnostic confidence for both the full- and half-count data. The levels of diagnostic confidence are shown in Figure 10.

Additional comparisons were performed on the full- and half-count stress-only data. There was no statistical difference for summed stress scores, ejection fraction or the need for resting images between the full-count and half-count data. In the population studied, there was a desire for resting data in approximately 20% of the images interpreted with stress-only images. These results are presented in Figure 11.
Figure 9 Comparison of the diagnostic accuracy of
full- and half-count data where both were reconstructed
using Astonish with both scatter and attenuation
corrections.Figure 10 The interpretative certainty of stress-only
interpretations from images using full- and half-count
data that were both reconstructed using Astonish with
scatter and attenuation corrections.Figure 11 Comparison of the stress-only results that were derived from full- and half-count
images and reconstructed using Astonish with scatter and attenuation corrections.
Figure 9
Figure 10
Figure 11
Comparison of the diagnostic accuracy of full- and half-count data where both were reconstructed using Astonish with both scatter and attenuation corrections.
The interpretative certainty of stress-only interpretations from images using full- and half-count data that were both reconstructed using Astonish with scatter and attenuation corrections.
Comparison of the stress-only results that were derived from full- and half-count images and reconstructed using Astonish with scatter and attenuation corrections.

Finally, a comparison was performed between stress-only half-count Astonish with scatter and attenuation correction and traditional rest/stress FBP SPECT. The same diagnostic accuracy was found despite using only a single image set (stress-only) and with half the counts.

Conclusions
When images were reconstructed using Astonish without attenuation correction, this study demonstrated significant improvements in perfusion image quality for both full- and half-count data. Furthermore, there was no loss in diagnostic accuracy using coronary catheterization as the gold standard.

When full- and half-count data were reconstructed using Astonish with scatter and attenuation correction, this study demonstrated a significant increase in normalcy and specificity with no significant loss in sensitivity.

These data demonstrate that the use of Astonish technology with or without attenuation correction improves image quality and may be applied to reduced acquisition time studies without compromising diagnostic accuracy. The ability to reduce imaging times should lead to reduced patient discomfort, reduced likelihood of patient motion, and improved laboratory throughput and overall efficiency. Alternatively, absorbed radiation doses can be reduced by injecting less activity while imaging for conventional imaging times.

The stress-only results demonstrate that perfusion and gated imaging can be performed on half-count data with high diagnostic accuracy, acceptable image quality, high interpretive certainty, similar defect extent and severity to full-time acquisitions, and a low perceived need for rest imaging. Performing studies using stress-only increases patient acceptance, further reduces radiation doses (compared to stress/rest half dose imaging), and improves laboratory efficiency.

Acknowledgments
The authors would like to acknowledge all the hard work of done by Staci Courter and her team in coordinating and collating the data.

References
  1. Venero CV, Heller GV, Bateman TM, McGhie AI, Ahlberg AW, Katten D, et al. A multicenter evaluation of a new post-processing method with depth-dependent collimator resolution applied to full-time and half-time acquisitions without and with simultaneously acquired attenuation correction. J Nucl Cardiol 2009; 16(5): 714-725.
  2. Bateman TM, Heller GV, McGhie AI, Courter SA, Golub RA, Case JA, et al. Multicenter investigation comparing a highly efficient half-time stress-only attenuation correction approach against standard rest-stress Tc-99m SPECT imaging. J Nucl Cardiol 2009; 16(5): 726-735.
  3. Ye J, Song X, Zhao Z, Da Silva AJ, Wiener JS, Shao L. Iterative SPECT reconstruction using matched filtering for improved image quality. Nuclear Science Symposium 2006, Conference Record. IEEE 2006; 4: 2285-2287.
  4. Frey EC, Tsui BMW. A new method for modeling the spatially-variant, object-dependent scatter response function in SPECT. Nuclear Science Symposium, 1996. Conference Record. IEEE 1996; 2: 1082-1086.
  5. American Society of Nuclear Cardiology. Imaging Guidelines for Nuclear Cardiology Procedures. J Nucl Cardiol. 2006; 13(6): e25-e171.


This content has been made possible by NetForum Community.
Share this on: Share your link in twitter Share your link in facebook Share your link on LinkedIn Print Rate this article: Log in to vote

 
Rating:
Votes:
0
Views:
4446
Added:
Jul 14, 2010

Rate this:
Log in to vote
 

White Paper
Astonish, Cardiac, coronary angiography, coronary arteries, coronary artery disease, dose, ECG-gated, FBP reconstruction, image quality, iterative reconstruction, myocardial perfusion, stenosis, Vascular
 

Clinical News
Best Practices
Case Studies
Publications and Abstracts
White Papers
Web seminars and Presentations
ExamCards
Protocols
Application Tips and FAQ
Training
Try an Application
Business News
Case Studies
White Papers
Web Seminars and Presentations
Utilization Services
Contributing Professionals
Contributing Institutions
Become a Contributor