This article documents the clinical value of PET and how PET imaging data is enhanced by computer software. Software can help process and analyze the data and manage the department at every level. Our experience with PET software analysis reveals the remarkable reproducibility and accuracy of PET Rb-82 myocardial perfusion imaging. It is in the best interest of everyone performing PET Rb-82 imaging to make use of such software analysis to achieve greater uniformity of quality.--
This article documents the clinical value of PET and how PET
imaging data is enhanced by computer software. Software can help
process and analyze the data and manage the department at every
level. Our experience with PET software analysis reveals the
remarkable reproducibility and accuracy of PET Rb-82 myocardial
perfusion imaging. It is in the best interest of everyone
performing PET Rb-82 imaging to make use of such software analysis
to achieve greater uniformity of quality.--Randolph E. Patterson,
Impact of Computer Analysis on the Clinical Value of
Positron Emission Tomographic Myocardial Perfusion Imaging with
Positron emission tomographic (PET) myocardial perfusion imaging
(MPI) with rubidium-82 (Rb-82) offers a powerful approach to
cost-effective management of coronary artery disease (CAD).
This article will discuss how computer software can enhance the
value of PET MPI.
Technical Advantages of PET Rb-82
Several features of PET provide images that are far superior to
those obtained from single-photon emission computed tomography
(SPECT), including: simultaneity correction; better spatial
resolution (full width at half max [FWHM]) values of 4 to 10 mm,
rather than 15 to 20 mm with SPECT
; enhanced signal-to-noise (PET achieving count rates that are
about ten-fold higher than with SPECT); and correction for
attenuation (absorption or deflection of counts by the body of the
person lying in the camera, creating artifactual defects on SPECT
Although there are continuing efforts to compensate for
attenuation in SPECT, it is a much more difficult problem than in
PET, requiring many more assumptions.
Clinical trials of attenuation-corrected SPECT MPI have been
PET, on the other hand, offers very straightforward and
well-established attenuation correction.
In brief, a "blank" scan is performed by rotating a source of
positron emitting radionuclide around the scanning table, before
the patient is in the camera, and recording the number of counts
measured by each detector. Later, this procedure is repeated with
the patient lying on the table in the camera, producing an image
that resembles a CT scan of the chest ("transmission" scan; Figure
1). The data from the transmission and blank scans are then used to
correct the patient's PET Rb-82 myocardial perfusion images for the
degrading effects of attenuation.
The practical effect of this correction is to produce MPIs that are
free of distortions created by overlying structures in the chest.
PET Rb-82 myocardial perfusion images can be acquired, processed,
interpreted, and reported within 75 minutes.
Computer Software for Quantitative Analysis of PET Rb-82
Each PET manufacturer offers software packages for the analysis
of PET images from their scanner. There are also a variety of
programs available to supplement the manufacturers' software. In
this article, we report on our experience with PET analysis
software we've developed (INPET) to illustrate how PET image
analysis and quantification works. However, such analysis can be
performed with any similar software.
Software can analyze PET Rb-82 myocardial images, in terms of
relative counts in regions of the left ventricle (LV).
The short axis images of the left ventricle (LV) can be displayed
as polar maps, and these images can be summed in a group of normal
individuals to represent the average LV distribution of PET Rb-82,
as we have done previously for SPECT.
Then each individual patient could be compared with this average LV
count distribution, quantitatively, to express the differences from
average as standard deviations (SD) below or above the mean normal
value for that region or voxel. Enhanced resolution and attenuation
correction with PET suggests that quantitative software for PET MPI
can determine the extent and severity of perfusion defects more
accurately than when the same software is applied to SPECT MPI.
In looking at a group of patients, we found remarkable
advantages of PET Rb-82 for MPI acquired at rest and with
dipyridamole-induced vasodilation ("stress").
We compared the spatial heterogeneity of MPI in subjects with low
clinical probability of CAD: 25 men and 25 women for PET Rb-82 and
25 men and 29 women for tread-mill stress Sestamibi and rest
The same software was used to generate average and standard
deviation polar maps (PMs) (Figure 2) from short-axis slices. For
quantitative comparisons, PMs were segmented into 13 regions. PET
Rb-82 MPI showed little heterogeneity with only small differences
between men and women. Stress Mibi PMs were more heterogeneous, and
virtually identical to rest Tl-201 PMs [
= NS for 13 segments, Sestamibi versus Tl-201 [men (M) and women
(W)]. Table 1 lists count ratios between different myocardial wall
In summary, PET MPI was uniform and eliminated the following
uncertainties caused by attenuation on SPECT: (a) differences
between women and men on stress MPI, (b) the inferior wall "defect"
in men, and the lateral wall "hot spot" in women and men. Of note,
there was no difference between rest Tl-201 and stress sestamibi
for SPECT in either women or men.
Also, our INPET software aids in the identification of fixed
versus reversible defects from Stress and rest PET Rb-82 myocardial
perfusion images. Defect areas and normal areas from stress are
mapped to the rest polar maps (SMAP polar display). Likewise,
defect areas from the rest polar map are mapped to the stress polar
map (RMAP). Finally, defect areas from rest are subtracted from the
corresponding defect areas at stress and mapped to the stress and
rest polar maps (S-R MAP) to indicate regions of "reversible
defects"abnormal on stress but normal on rest. Numerical data are
presented to quantify changes in defect size and severity, and the
change in the ratio of ischemic zone counts to normal zone counts
between stress and rest images. Figure 3 presents a patient study
illustrating the clinical value of using quantitative PET MPI
Clinical Studies Comparing PET MPI with Coronary
Angiography Using PET MPI software
Churchwell et al,
in our group, compared quantitative PET Rb-82 MPI data with
clinical data (n = 52) or coronary angiography (n = 91) to
determine whether or not patients had CAD. Clinical data, including
history of symptoms and risk factors and resting electrocardiogram
findings, were used to identify a group with a very low probability
of CAD, based on Framingham data
for our normal file. Coronary angiography results were analyzed by
computer to define the percentage diameter stenosis.
For the purpose of this study, CAD was defined as >40% reduction
in lumen diameter; 72 of 91 patients had one or more arteries with
lesions that met these criteria.
We analyzed PET MPI by varying the threshold values for defect
size (% LV) and severity (average SD below mean of normal file) and
found by this receiver-operator characteristic curve analysis
that the best results to define an abnormal stress PET are:
containing defect(s) of >=5% LV, with Rb-82 activity >=2.5 SD
Interobserver Agreement in PET MPI
One of the most striking results was the remarkable
interobserver agreement among all three interpreting physicians in
94% of 142 patients, with 95% confidence intervals (CI ) of 90% to
97%; and between two of the physicians in 97% of cases (CI = 95% to
Our group had previously found a lower 89% (CI = 83% to 94%)
agreement between two observers with SPECT, even though the
quantitative software was similar.
Such a high level of interobserver agreement with interpretation of
PET MPI was a direct benefit of using quantitative software, and
has major implications for the clinical usefulness of the test.
Comparison of PET MPI with Coronary Angiography
Using the same criteria for CAD, 71 of 72 patients with
angiographic CAD were identified with PET MPI (99% sensitivity, CI
= 96% to 100 %). Sixteen of 19 patients with no evidence of CAD on
angiography were identified as "normal" (84% specificity, CI = 67%
to 100%) (Table 2).
Impact of "Referral Bias" on PET/Angiography
Identification of patients as normal by coronary angiography
creates a major disadvantage in that the people had some
significant symptoms or disease that led to the referral to the
cardiac catheterization laboratory.
This problem of referral bias tends to select a small minority of
people with normal test results for cardiac catheterization, so
that calculated specificity is reduced artificially.
Conversely, referral bias also means that only a small minority of
people with normal test results are referred for cardiac
catheterization, so that calculated sensitivity is inflated.
One way to understand this issue is to study test results in people
with a low probability of CAD based on clinical information.
Thus, with PET MPI evaluation of 52 people with very low clinical
probability of CAD, we found that all of them were normal (100%
Impact of PET on Uncertainties in Interpretation of
In one year, 1997, we documented another benefit of PET. In 2748
patients there was a dramatic reduction in the number of
interpretations by two experienced physicians that were classified
as "probably" instead of "definitely" normal or abnormal; from 37%
with SPECT, to 21% with PET (Table 3). The possibility of a
nondiagnostic or uncertain interpretation of a noninvasive test has
been one of the major factors driving physicians to recommend more
costly invasive diagnostic studies. These results indicate that PET
can produce unequivocal results in the great majority of
PET Rb-82 offers a remarkably reliable, accurate assessment of
myocardial perfusion that can improve clinical practice of
cardiology. This article demonstrates how quantitative analysis of
the images and comparison to normal files can enhance the value of
the information of PET data.
The benefits of this analysis include: enhanced reproducibility of
interpretations among different physicians; enhanced accuracy of
discriminating whether or not a patient has CAD; increased
certainty of diagnosis; rapid generation of reports that show
uniformity; and creation of a database that allows comparison of
clinical factors with quantitative image analysis.
17. For more information on the software, see our Web site: