QGS ejection fraction reproducibility in gated SPECT comparing pre-filtered and post-filtered reconstruction

Introduction Gated myocardial perfusion single photon emission computed tomography (SPECT) was developed in the late 1980s with the initial role of differentiating artefact from pathology [1]. The high count density of 99mTc based myocardial perfusion studies allows SPECT images to be synchronized to the patient’s electrocardiogram (gated) because they maintain adequate count density in individual cardiac frames (intervals). This feature, in conjunction with acceptable spatial and contrast resolution, allows simultaneous assessment of myocardial perfusion and ventricular function [2].
Despite limitations associated with filtering gated SPECT data, it is universally recommended that default filter parameters are adhered to due to the danger of introducing false positive or false negative results following filter customization [3]. Over-filtering myocardial perfusion SPECT data is known to cause false negative results and under-filtering causes false positive results [3]. Since the quantitative gated SPECT (QGS) software (Cedars Sinai Medical Centre, Los Angeles, California) determines functional parameters utilizing Kubo et al. [4] indicated that the ideal gated SPECT prefilter for ventricular volume determination was a Butterworth with a critical frequency of 0.54cycle/cm but this filter caused an underestimation of left ventricular volumes by 17% for end diastolic and 8% for end systolic. Kubo et al. [4] conclude that pre-filters have little impact on the ejection fraction because of the simultaneous blur effect on end diastolic volume (EDV) and end systolic volume (ESV), effectively cancelling each other out. Fredericks et al. [5] found that QGS was robust to both variations in filter cut-off and orientation of the short axis. The authors of QGS recommend a Butterworth pre-filter of 0.3cycle/pixel at a pixel size of 0.64cm [6] which translates to a recommended critical frequency of 0.47cycle/cm. This recommendation is discordant with the default filter prescribed by the QGS software manual [7] which requires a three-dimensional low pass postfilter with an order of 5.0 and a cut-off of 0.21. Not surprisingly then, 60.4% (32/53) of departments using QGS also employ a pre-filter [8]. This may be attributed in part to the requirement of pre-filtering in ungated quantitative software (e.g., CEqual), which was used widely prior to the widespread use of gating, and to the discordant recommendations in the literature.

The aim of this investigation was to determine the impact on the QGS determined functional parameters by comparing pre-filtering to post-filtering in the gated myocardial perfusion SPECT reconstruction process.

Methodology This clinical investigation employed a retrospective repeat-measures design. This study design allowed assessment of the dependant variables (ejection fraction, EDV, ESV) by manipulating the independent variable (reconstruction strategy) for a single data set. Using this approach meant that a single clinical data set acted as both the control group (post-filtering) and the experimental group (pre-filtering).

All patients in the study population followed one of two myocardial perfusion SPECT protocols: 2 day rest–stress or 2 day stress–rest. All myocardial perfusion SPECT studies employed a 740MBq (20mCi) dose of 99mTc tetrofosmin (Nycomed-Amersham, Amsterdam). A triple detector gantry was used to acquire all patient data. All data acquisitions employed low energy, high resolution collimation with step and shoot mode, elliptical orbits and a 64 matrix. The zoom was 1.23 and projections were acquired at 31 intervals for 20s per projection to provide a total acquisition time of 15min. All patients were positioned supine with their feet into the gantry for an eight interval gated SPECT acquisition. A total of 25 patient files were examined, each with both a gated rest and gated stress study and, thus, a total of 50 studies were produced for quantitative analysis with QGS.

The gated SPECT data for all clinical studies were reconstructed using two strategies. The first employed pre-filtering with a Butterworth low pass filter (order 4.0 and cut-off 0.21) followed by reconstruction and reorientation of gated short axis slices for QGS analysis. The second strategy employed post-filtering with a Butterworth low pass filter (order 5.0 and cut-off 0.21) during reconstruction followed by re-orientation of gated short axis slices for QGS analysis. All data was reconstructed using a 1801 filtered back-projection algorithm.

The statistical significance was calculated using the Student’s t-test for continuous data. A P value less than 0.05 was considered significant. Normality of distribution was determined using the Shapiro–Wilk W test with a P value less than 0.05 indicating that the data varies significantly from normality. The differences between independent means and proportions was calculated with a 95% confidence interval (CI). Correlation was evaluated with chi-squared analysis and reliability measured using Cohen’s kappa coefficient. Bland–Altman analysis [9] and the matched pairs t-test were used to assess agreement between paired data.

Approval for this study was granted by the Charles Sturt University Ethics in Human Research Committee for the retrospective manipulation of the de-identified patient data.

Results All 25 clinical studies had both stress and rest data quantified with QGS software following reconstruction by both a pre-filtering algorithm and a post-filtering algorithm by two experienced observers. Inter-operator reproducibility was excellent with a correlation coefficient of 0.994 for post-filtered data and 0.979 for prefiltered data (R2 of 0.988 and 0.958, respectively). No statistically significant difference was noted between matched pairs comparing operators 1 and 2 for either the pre-filtered data (P=1.0) or the post-filtered data (P=0.06). Consequently, the following analysis was undertaken using the mean data of matched pairs between operators.

The mean ejection fraction for the post-filtered data was 49.5% (95% CI, 45.8–53.1%) and for the pre-filtered data was 54.8% (95% CI, 51.4–58.1%). Excellent correlation was demonstrated between the pre- and post-filtered ejection fractions with a correlation coefficient of 0.964 and R2 of 0.929 (Fig. 1). While this supports a strong relationship between the pre- and post-filtered ejection fractions, it does not provide an indication of the strength of agreement between data. Consequently, Bland–Altman analysis was performed (Fig. 2) which indicated that the 95% limits of agreement included 96% of data points. The difference between matched pairs of pre-filtered and post-filtered ejection fractions demonstrated a normal distribution (P=0.19). The mean difference between matched pairs of pre- and post-filtered ejection fraction data was 5.3% (95% CI, 4.3–6.3%) where a positive difference indicates that the pre-filtered ejection fraction is higher than that of the post-filtered. Despite the overlap of 95% CIs, the match pair t-test demonstrated a statistically significant difference between matched pairs (P<0.0001) and a statistically significant difference was shown between the means (P=0.005).

Bivariate analysis of the ejection fraction (EF) calculated following pre-filtered reconstruction versus post-filtered reconstruction demonstrating excellent correlation (0.964).
Bland–Altman analysis of the pre-filtered ejection fraction and the postfiltered ejection fraction demonstrating a mean difference of 5.3% (solid horizontal line).

The pre-filtered mean EDV was 106.6ml and the postfiltered was 103.2ml. Despite excellent correlation (correlation coefficient=0.967), a statistically significant difference was noted between matched pairs (P<0.0001). The mean difference between pre- and post-filtered EDV was –3.38ml indicating that the prefilter EDV was lower than the post-filtered. Similarly, the pre-filtered mean ESV was 57.7 ml and the post-filtered was 50.8. Despite a correlation coefficient of 0.974, a statistically significant difference was noted between matched pairs (P<0.0001). The mean difference between pre- and post-filtered ESV was –6.95ml. The greater impact of filtering on ESV than EDV (absolute and relative) is responsible for the ejection fraction differences described above.

In Conclusion

It is worth noting that using an eight-bin gated acquisition underestimates the ejection fraction by 3.7% [10] requiring interpreting physicians to add 4% to the calculated ejection fraction [7]. An eight-bin gated SPECT acquisition processed using a pre-filter would not require this correction because it overestimates the ejection fraction by 5.3% compared to the prescribed post-filter. The uncorrected result, therefore, would leave the ejection fraction with a 1.6% overestimation of the ejection fraction. It is clear that accurate interpretation and reporting of functional parameters on gated myocardial perfusion SPECT using QGS requires either adherence to the prescribed acquisition and processing parameters or a thorough understanding of the implications of variations to these prescribed parameters.

In general, pre-filtering in SPECT reconstruction with a ramp filter is thought to be superior to post-filtering. Armed with this knowledge and combined with variations in recommendations offered in the literature, discordance between clinical practice and the recommendations of the software distributor is not uncommon. Nonetheless, the default filtering requirements for QGS software includes post-filtering of reconstructed data. The impact of performing pre-filtering on data in the reconstruction process is significant with a 5.3% increase in the calculated ejection fraction over post-filtering. Clearly, this has the potential to undermine diagnostic and prognostic roles of functional parameters. Furthermore, precision and reliability of functional parameters are undermined, especially between centres employing different filtering strategies.



Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store