CARDIOVASCULAR JOURNAL OF AFRICA • Volume 25, No 2, March/April 2014
48
AFRICA
This was also found in our study, which showed the Siemens
software gave higher estimates of LVEF than the Hermes
software. Hiscock
et al
,
3
who compared 18 workstation algorithm
combinations including Siemens and Hermes, also found that
the Siemens software gave estimates of LVEFs that were higher
than estimates obtained from the Hermes software. The mean of
64 estimates of LVEF calculated with the Siemens software was
64.6% and with the Hermes software 61.3 and 61.7%.
Differences in estimates of LVEF obtained from different
software methods on the same patient could be attributed to
variations in the algorithms used for edge detection in determining
the ROI around the left ventricle and background subtraction.
There was no documentation on the Siemens package available
to us in the online user manual. There was limited documentation
on the mathematical algorithms used in the Hermes software
for determination of the left ventricular ROI and background
subtraction. The Hermes software package used a variation on a
second differential method for edge detection in determination of
the left ventricular ROI. Background subtraction was performed
in the background ROI outside the left ventricular ROI at the
end of diastole, as well as on the non-ventricular counts within
the end-diastolic region at end-systole. This could have resulted
in an over-subtraction of background and account for a slightly
lower LVEF.
In the studies performed by Skrypniuk
4
and Fair
et al
.,
5
it is
suggested that there is a need for software suppliers to supply
more information on their software packages and to give
guidance on data quality requirements as well as on any limits of
operation. Fair
et al.
5
suggested that adequate testing of software
packages against phantoms (if possible), and clinical testing on
a reasonable number of patients should be done by software
manufacturers.
Because different software packages use different algorithms
and give different values for LVEF, all reports of LVEF calculated
from GBP studies should contain the name and version of the
software package used to calculate the result. Wherever possible
the same software should be used to process serial studies. When
the same software is not used to process serial studies, ideally, all
the raw data should be retrieved from an archive and reprocessed
using the current software and a summary of all previous results
included in the current report. If this is not possible, the mean
difference and limits of agreement of the two software methods
should be given.
The pattern of our results for reproducibility is consistent with
previous reports in the literature by van Royen
et al
.,
11
Pfistererer
et al
.,
12
, Hains
et al
.,
13
Hiscock
et al
.,
3
and Skrypniuk
et al
.
4
Van Royen
et al
.
11
found that repeat quantitative radionuclide
assessments of LVEF can be expected to be within a 2–4% range
if a study is processed twice by the same operator. Pfisterer
et
al
.
12
do not state how many times each study was processed, but
found studies reprocessed by the same operator to be within a
1–3% range of each other and within a 1.4–5% range of each
other if processed by different operators.
We found that studies processed three times by the same
operator were within a 3–6% range of each other for the Siemens
method, and within a 3–9% range for the Hermes method. The
reason for the wider range in our study is most probably due to
the fact that our studies were processed three times. Two of our
three estimates were always more closely related. The difference
between the two closest estimates was
≤
3% in all patients with
both methods.
Our study is in agreement with that done by Hains
et al
.
13
were
the SD of the difference between estimates of LVEFs calculated
by one operator ranged from 0.5–0.8. Hiscock
et al
.,
3
who also
compared the Hermes and Siemens systems, reported the SDs of
the difference between LVEFs calculated by one operator to be
1.80 and 2.56 for Hermes software systems. Their reported value
for inter-operator variability for the Siemens software system,
4.79, was however much higher than the values of 0.6, 0.7 and
0.5 reported in our study for the three operators, respectively.
The reason for the higher value is not known. Skrypnuik
et al
.
4
reported the SD of the difference between LVEFs calculated by
one operator to be 0.002.
To date no articles on the influence of the acquisition
of GBP studies on different cameras could be found in the
English literature. In a study on the intra- and inter-observer
reproducibility of LVEF estimates obtained from gated
myocardial perfusion SPECT imaging and those obtained from
GBP imaging, Castell-Conesa
et al
.
14
collected data from two
institutions, which acquired their data on different cameras. The
issue of whether the acquisition of studies on different cameras
had an effect on results was however not specifically addressed.
Our study showed no difference in the results of the GBP studies
acquired on different cameras.
We suggest that each GBP study should be processed three
times before reporting a result. In our study, two of the three
LVEF estimates were closely related regardless of the software
method used. The difference between the two closest estimates
was always
≤
3%. The mean of these two estimates would
probably be the best representation of the patient’s LVEF.
Anthracyclines have been used for the past 30 years in
chemotherapy regimes. No consensus however exists on the
optimal monitoring for cardiotoxic effects. Guidelines have been
proposed, but none incorporate all of the necessary components
of monitoring for chemotherapy-induced cardiotoxicity.
15
Until such research is available, following one of the existing
guidelines is the most practical solution.
At our institution the oncologists use the guidelines as set out
in the
Oxford Textbook of Oncology.
16
This guideline suggests
that in patients with baseline LVEF estimates greater than 50%,
doxorubicin treatment should be discontinued if the LVEF
decreases by 10%, or if a value of 50% is reached. In patients
with baseline LVEFs of less than 50% but greater than 30%, it is
suggested that doxorubicin therapy should be discontinued if the
LVEF decreases by 10% or if a value of less than 30% is reached.
Skrypniuk
et al
.
4
found the change in a LVEF value required
to be 95% confident of a real change when carrying out repeat
measurements on an individual patient to be 4.5%. It is therefore
recommended that each department obtain their own percentage
values for clinical decision-making for each software package
used independently.
In our study, the patients who had SDs for the three estimated
LVEFs above the 95th percentile had differences between the
highest and lowest (maximum difference) of the three estimates
of 6% using the Siemens software method, and 9% using
the Hermes software method. This implies that within our
department, a reduction in the ejection fraction for a follow-up
study of more than 6% is of clinical significance if processed
using the Siemens software method, and 9% when using the
Hermes software method. If the closest two of the three estimates
