Key words
liver - CT - image manipulation/reconstruction - segmentation - surgery - tissue
characterization
Introduction
Liver volumetry is used to predict a patient’s liver volume by the use of imaging,
usually with an abdominal CT scan. It is often used in the setting of living-donor
liver transplantation to predict the size of the future graft and to ensure a certain
volume of the remaining donor liver [1]
[2]
[3]
[4]
[5]. Liver volumetry is also important in patients undergoing extended liver resection.
In both cases a minimal amount of functional liver is necessary to maintain sufficient
metabolic and detoxifying function. Otherwise it can lead to a threating scenario
with acute liver failure, coagulopathy and multiorgan dysfunction [6].
Previous studies found a good correlation (p < 0.001) between liver volumes measured
by liver volumetry with the help of a CT scan and the real liver volume, measured
by water displacement [3]
[7]
[8]. Most of these studies evaluated their software in donors for living-donor liver
transplantations. Patients undergoing liver resection have been studied less frequently
[9]
[10] but are equally important in the clinical routine.
Manual liver volumetry was shown to be a time-consuming procedure. Nakayama et al.
measured 32.8 ± 6.9 minutes for one manual volumetry [8]. Automated liver volumetry programs promised to be faster. For example, Suzuki et
al. found a mean time of 0.57 ± 0.06 minutes with their semiautomated software [11].
In the daily routine at our clinic, liver volumetry is performed with the help of
a manual program, which gives the observer a lot of freedom in the definition of liver
tissue and segments to be resected. Volumetry is performed by different observers
with different experience levels. Although new observers are always introduced to
the program and the volumetry, considerable variation is still possible in the definition
of liver segments. This leads to intra- and interindividual differences. Semiautomated
software for the detection and segmentation of liver tissue could be of great help
in terms of a time reduction and volumetric measurement reproducibility. Several companies
are working on such software programs and our department had the opportunity to test
a prototype software, which is now also commercially available (CT Liver Analysis,
Philips Healthcare, Best, NL).
A study was performed to compare the manual program for liver volumetry with a prototype
version of the semiautomated software tool. The hypothesis was that the semiautomated
software would be faster, more accurate and less dependent on the evaluator’s experience.
Materials and Methods
Written informed consent was obtained from all patients for this prospective IRB-approved
study.
10 patients were included in this prospective study (4 women, 6 men), who underwent
hemihepatectomy in our Department for Visceral Surgery. The mean age was 70 years
(range: 56 – 83 years). The sample size was estimated to be high enough by our department
for biostatistics in view of the expected differences between the methods. Surgery
took place between 01/2012 and 01/2013. Underlying diagnoses were adenocarcinoma of
the gall bladder (n = 1), cholangiocellular carcinoma (n = 1), liver abscess (n = 1),
hepatocellular carcinoma (n = 2) and liver metastases (n = 5).
The resected parts of the liver had different sizes in each case. The resected Couinaud
segments were V-VIII (n = 3), IV-VIII (n = 1), I-IV (n = 2), II+III (n = 2), IV-VIII+I
(n = 1), V-VIII+IVa (n = 1).
All patients underwent a preoperative abdominal CT scan which is part of the normal
preoperative workup. The scans were performed on a 64-slice CT scanner (Philips Brilliance
64, Philips Healthcare, Best, NL) with 120 kV and 200 effective mAs, a collimation
of 64 × 0.625 mm and a reconstructed slice thickness of 3 mm. All patients were scanned
with 3-phase CT (unenhanced, arterial phase, portal venous phase; timing with bolus
tracking) after intravenous application of contrast material (0.5 g iodine per kilogram
body weight; Imeron 400 mCT, Bracco Imaging, Konstanz, Germany). The delay for the
start of the arterial phase scan and the portal-venous phase scan was set at 15 and
45 seconds, respectively, after a threshold of 100 Hounsfield Units was reached in
the abdominal aorta. Image data of the portal-venous phase were used for volumetry.
As a gold standard, the resected parts of the liver were measured with water displacement
volumetry immediately after resection in the operating room. Therefore, the resected
piece of liver was placed into a basin completely filled with NaCl solution and the
displaced fluid was collected and measured with a 1000 ml measuring cylinder.
Thereafter a CT scan of the resected liver, inserted in a box filled with formalin,
was performed on the same CT scanner with 120 kV, 100 effective mAs and 3 mm reconstructed
slice thickness. Volumetry was performed with two different types of software. The
first one is part of the CT viewer integrated into the dedicated CT workstation (Extended
Brilliance Workspace, Philips Healthcare, Best, NL) and routinely used in our department.
Here, the borders of the liver must be defined per slice by the user. Slice interpolation
is possible, but the user has to control the correct borders. The resulting volume
is displayed in milliliters. To measure parts of the liver, the user must exclude
the unwanted anatomy by hand. This can be done on the volumetric image of the liver
([Fig. 1]).
Fig. 1 The manual software (Extended Brilliance Workspace, Philips Healthcare, Best, NL).
Here the previously determined total liver volume is already divided into the resected
liver and the remaining part. The upper right picture shows the 3 D reconstruction.
Abb. 1 Die manuelle Software (Extended Brilliance Workspace, Philips Healthcare, Best, NL).
Die vorher volumetrierte Gesamtleber wurde hier bereits in das zukünftige Resektat
und den verbleibenden Leberteil untergliedert. Das obere rechte Bild zeigt die 3D-Rekonstruktion.
The second software was a prototype software for liver volumetry and segmentation
developed by Philips which is now commercially available (“CT Liver Analysis”, Philips
Healthcare, Best, NL). Here, the liver volume is identified automatically, but the
borders can be corrected by the user. In a second step, vessels are automatically
identified and classified as portal, hepatic, or unclassified veins. Their volume
is included in the total volume. Manual correction is possible. In a third step, for
segment definition, the user must set nine points (bifurcation of right portal vein,
vena cava inferior, right hepatic vein, mid hepatic vein, umbilical fissure, superficial
and deep ligamentum venosum, end of left portal vein, left tip of the liver) from
which Couinaud segments are calculated automatically. The volumes for every single
segment are listed and can be accumulated to the respective volumes of the resected
liver part ([Fig. 2]).
Fig. 2 The semiautomated software (CT Liver Analysis, Philips Healthcare, Best, NL). The
liver is already divided into the different Couinaud segments. In the upper left picture
the 3 D reconstruction is shown. The schematic drawing of the liver on the left side
indicates where the landmark points must be set to determine the Couinaud segments.
Abb. 2 Die halbautomatische Software (CT Liver Analysis, Philips Healthcare, Best, NL).
Die Leber wurde bereits in die Segmente nach Couinaud unterteilt. Im oberen linken
Bild ist die 3D-Rekonstruktion zu sehen. Die links sichtbare schematische Darstellung
der Leber zeigt wo die Landmarken gesetzt werden müssen, die die Couinaud-Segmente
definieren.
Volumetry of the preoperative CT dataset was performed by six users. Three of them
were radiologists experienced in the use of the manual software and three of them
were medical students and novices in CT volumetry. The radiologists had three, four
and fourteen years of clinical experience and the medical students were in their fourth
and fifth year of study. They all received the same introduction to the prototype
software. Medical students received an additional introduction to the manual program.
Each user performed test volumetry on three different datasets to become accustomed
to the programs and to minimize a learning effect during the study. Thereafter, each
user performed volumetry of the whole liver and of the resected segments with the
manual program as well as with the semiautomated software for each of the ten livers.
The gallbladder was excluded and tumors were included in the total liver volume. Resected
segments were known from the operating room report and were made available to the
users. In contrast to the approach in the clinical routine, the preoperative volumetry
was also done after the time of surgery. For each method the measurement was repeated
three times in a random order to avoid a learning effect. Time needed for volumetry
and measured volumes were noted and analyzed separately for the two groups of observers.
Volumetry of the postoperative CT dataset of resected liver parts was performed by
three users, two radiologists and one medical student. All of them had also done the
preoperative measurements. This measurement was done as a quality check for accuracy
of both volumetry modalities and not to assess differences between observers. The
manual and semiautomated delineation of the resected liver part proved to be simple,
because the liver and the circumfluent formalin had very different Hounsfield Units.
This is why only three participants also did the postoperative measurements.
Statistical Analyses
Statistical analyses were done with SPSS for Windows Version 15.0 and performed in
cooperation with the local Institute for Medical Informatics, Statistics and Epidemiology.
Descriptive statistics including mean value, standard deviation and range were performed
for measured times and volumes, in each case separately for resected and total liver.
The results of measured volumes were visualized with Bland-Altmann plots [12]. For correlations the spearman correlation coefficient (-1 ≤ρ≤ 1) was used. Statistical
tests to compare measured volumes were the Wilcoxon test and the Mann-Whitney U-test.
For the analysis of the experience level, the use of Cohen’s Kappa was not recommended
by our institute for statistics. Instead visualization with Bland-Altman plot was
recommended.
P-values ≤ 0.05 were considered significant.
Results
An overview of all results can be seen in [Table 1].
Table 1
Mean values of all measured times [minutes:seconds] and volumes [ml].
Tab. 1 Mittelwerte aller gemessenen Zeiten [Minuten:Sekunden] und Volumina [ml].
|
mean value measured by radiologists
|
mean value measured by medical students
|
mean value over all observers
|
|
time needed to measure the total liver with the manual method [minutes:seconds]
|
5:00 ± 1:51
|
8:59 ± 2:45
|
6:59 ± 3:04
|
|
time needed to measure the total liver with the semiautomated software
[minutes:seconds]
|
1:53 ± 1:13
|
1:42 ± 1:08
|
1:47 ± 1:11
|
|
time needed to measure the resected liver with the manual method
[minutes:seconds]
|
2:02 ± 1:09
|
2:35 ± 1:53
|
2:19 ± 1:35
|
|
time needed to measure the resected liver with the semiautomated software
[minutes:seconds]
|
4:16 ± 2:09
|
4:50 ± 2:07
|
4:32 ± 2:09
|
|
total volume measured preoperative with the manual method [ml]
|
1954 ± 667
|
2126 ± 707
|
2040 ± 691
|
|
total volume measured preoperative with the semiautomated software [ml]
|
1818 ± 572
|
1860 ± 579
|
1839 ± 574
|
|
resected volume measured preoperative with the manual method [ml]
|
1084 ± 764
|
1199 ± 785
|
1141 ± 775
|
|
resected volume measured preoperative with the semiautomated software [ml]
|
970 ± 625
|
1006 ± 695
|
988 ± 659
|
|
water displacement volumetry [ml]
|
|
|
837 ± 740
|
|
resected volume measured postoperative with the manual method [ml]
|
|
|
795 ± 603
|
|
resected volume measured postoperative with the semiautomated software [ml]
|
|
|
809 ± 620
|
Liver Volume
In each case the mean volume of the total and the resected liver, measured on the
preoperative dataset with the manual method, was greater than the volume measured
preoperatively with the semiautomated software. The differences, both for the total
and the resected volume, were statistically significant (p < 0.001).
Postoperative measurement of resected segments with CT scan shows results very close
to water displacement volumetry with both methods. Differences are not statistically
significant (manual: p = 0.557; semiautomated: p = 1.000). Correlations to the gold
standard were very strong with both methods (manual: ρ = 0.988; semiautomated: ρ = 0.988).
Comparison of volumes measured preoperatively with volumes measured by water displacement
shows that the manually predicted volume is always greater than the volume measured
by water displacement (mean difference 305 ml (57 %); correlation ρ = 0.927). The
volume predicted with the semiautomated software as compared to the water displacement
is bigger in some cases and smaller in others with an excellent correlation (mean
difference 152 ml (33 %); correlation ρ = 0.939) ([Fig. 3]). Both differences are statistically significant (manual: p = 0.002; semiautomated:
p = 0.027).
Fig. 3 Bland-Altman plot (lines show mean of the difference, mean of the difference plus
1.96SD and minus 1.96SD) for differences between preoperative CT volumetry and water
displacement volumetry: a manual volumetry, b semiautomated volumetry.
Abb. 3 Bland-Altman-Plot (Die Linien zeigen den Mittelwert der Differenz, sowie den Mittelwert der
Differenz plus bzw. minus 1,96 × Standardabweichung) des Unterschieds zwischen der
präoperativen CT-Volumetrie und der Verdrängungsvolumetrie für die: a manuelle Volumetrie, b halbautomatische Volumetrie.
Measurement time
The required mean time to determine total liver volume with the manual method was
6:59 ± 3:04 minutes (range: 2 – 18 minutes). With the semiautomated software the mean
time was 1:47 ± 1:11 minutes (range: 1 – 8 minutes). So the semiautomated software
is 3.9 times faster than the manual measurement. The difference is statistically significant
(p < 0.001).
Measurement of the preoperative volume of resected liver parts took a mean time of
2:19 ± 1:35 minutes with the manual method and 4:32 ± 2:09 minutes with the semiautomated
software. For the combination of both measurements, the semiautomated software is
1.5 times faster than manual volumetry (manual: 9:22 ± 3:09 minutes, semiautomated
6:20 ± 1:40 minutes). This difference is also statistically significant (p < 0.001).
Experience level
In the measurement of the total liver volume, the absolute difference between medical
students and radiologists was on average smaller with the semiautomated software (41 ml
with semiautomated software versus 171 ml manually measured), but the difference was
not statistically significant (manual: p = 0.208; semiautomated: p = 0.722).
In the measurement of the resected liver on preoperative scans, the differences between
the two groups of users were also not statistically significant in both methods (manual:
p = 0.300; semiautomated: p = 0.912) ([Fig. 4]).
Fig. 4 Bland-Altman plot (lines show mean of the difference, mean of the difference plus
1.96SD and minus 1.96SD) for differences between the measurement of the resected liver
by students versus radiologist with: a manual volumetry, b semiautomated volumetry.
Abb. 4 Bland-Altman-Plot (Die Linien zeigen den Mittelwert der Differenz, sowie den Mittelwert
der Differenz plus bzw. minus 1.96 × Standardabweichung) des Unterschieds zwischen
den Resektatvolumetrien durch die Studenten beziehungsweise durch die Radiologen mit
der: a manuellen Volumetrie, b halbautomatischen Volumetrie.
Comparing the results of the resected liver volume on preoperative scans with the
real volume of the resected liver, radiologists had smaller differences with both
methods (mean difference 305 ml manually and 134 ml with semiautomated software) than
medical students (mean difference 362 ml manually and 170 ml with semiautomated software).
The differences are statistically significant (radiologists manual: p = 0.002; medical
students manual: p = 0.002; medical students semiautomated: p = 0.014) except for
the difference between the volume measured by the radiologists with the semiautomated
software and the real volume (p = 0.084).
Radiologists were faster in measuring the total liver and the resected volume with
both methods, but the differences were not statistically significant (manual: p = 0.259;
semiautomated: p = 0.125).
Discussion
Total Liver Volume
Heinemann et al. measured volumes of 33 healthy livers post mortem by water displacement
and found a mean liver volume of 1862 ml for the Caucasian population [13].
The present study provides results comparable to those of Heinemann. Suzuki et al.
found a manually measured mean liver volume of 1486 ± 343 ml (DICOM viewer, Abras
version 0.9.9) and an interactively measured volume of 1520 ± 378 ml (Volume Tracing
in Advanced Vessel Analysis, Philips Healthcare) [11]. They used a similar software type for volume tracing as we did for manual volumetry,
but with another tool for liver delineation. Their mean liver volume was 434 ml smaller
than the volume found in the present study which might be due to different measurement
tools.
D'Onofrio et al. used the same semiautomated software as we did but compared it to
software running on a personal computer. They found a total liver volume of 1787.31 ml,
which is comparable to the results found in the present study [14]. Certainly every method has its own limitations and it is difficult to compare absolute
volumes across different studies.
A general problem that can also affect the volume measurement is the slice thickness.
We used slices with a thickness of 3 mm, because this is routine in our clinic and
both programs can deal with this thickness. Hori et al. reported that if a maximum
error of 5 % is allowed, a slice thickness of 5 mm is exact enough and smaller slices
make volumetry even more accurate. Thinner slices can additionally lead to the problem
that the volumetry is more time-consuming [15].
Resected Liver Volume
So far water displacement is the best method to measure the real volume of resected
liver parts [16] and better than measuring liver weight because density of liver tissue varies. Lemke
et al. found densities between 0.67 g/cm³ and 1.66 g/cm³ [17]. Volumes measured by water displacement in the present study cannot be compared
to volumes found in the literature, because we only measured resected liver parts
which differ a lot in size, because different segments were resected in every case.
However, D'Onofiro et al. also measured resected liver parts with different sizes
with the same semiautomated software as used in the present study. They found a mean
volume of resected liver parts of 1021.23 ml, which is comparable to our results,
although they only used the weight of the surgical specimen as the gold standard [14]. Additionally, they did not give any information about the time needed for volumetry
and reproducibility.
In our study both methods resulted in an overestimation of the real volume of resected
liver parts in almost all cases. This overestimation was also found by Lemke et al.
[2], who compared the weight of right liver lobes preoperatively in vivo and after resection
and found a correction factor of 0.75. Niehues et al. also found a 13 % overestimation
of in vivo volumetry in a pig model [18]. They found that intraoperative blood loss is the main reason for overestimation.
Hwang et al. found a difference between CT measurement and blood-free water displacement
volumetry of 20 % but only a difference of 4 % between blood-filled graft volume and
CT volumetry [19]. In view of these results, overestimation of liver volume in the present study can
be fully explained.
Additionally, comparison of CT volumetry of explanted liver parts and displacement
volumetry did not suffer from this fluid problem and showed an excellent correlation
with both methods.
Measurement time
Nakayama et al. measured 32.8 ± 6.9 minutes with their manual program and 4.4 ± 1.9
minutes with their automatic program [8]. Suzuki et al. found a mean time of 39.4 ± 5.5 minutes for manual volumetry and
0.57 ± 0.06 minutes for semiautomated software. With their “interactive software”,
which was similar to our manual software, they needed 27.4 ± 4.6 minutes. This is
much longer than the mean time per case found in the present study and might be explained
by the use of another tool within the software [11]. In comparison with these studies the manual software tested in the present study
seems to be considerably faster, even in unexperienced users. This can be explained
by use of the interpolation mode, we used in this study, in which it is not necessary
to define liver borders in every single CT slice. The time needed for semiautomated
volumetry in the present study is comparable to the time found in literature. The
time needed for measurement of resected parts was found to be a bit longer with the
semiautomated software in the present study as compared to manual volumetry. The reason
is the time that was needed to set the nine points to define the segments of the liver
correctly. However, with the semiautomated software volumes of all liver segments
and vessels are calculated within this time. With the manual volumetry it is only
possible to get the volume of the liver part the user detaches from the formerly measured
total volume.
After all, the manual volumetry method used in the present study is already relatively
fast, but the tested software is up to four times faster. In comparison to other programs
discussed in the literature, the time gain can be even more substantial.
Experience level
In the abovementioned studies [8]
[11]
[14], just one user measured all volumes. Even if the radiologist is experienced, this
can lead to a bias. Correct and repeatable results are also very important in the
clinical routine. Frericks et al. performed a study with three equally trained observers,
who re-measured livers in complicated cases only. The three observers worked together
to measure one volume for one liver and no comparison of different results was made
[7]. Radtke et al. had three observers (one trained radiologist and two untrained surgeons),
but they also did not compare different results [1]. Sandrasegaran et al. performed a study to expose reproducibility and interobserver
variations in liver volumetry, but they had just 2 observers and did not provide any
information about their experience level. Additionally just one observer repeated
the measurement. They found extremely high inter- and intraobserver correlations (r = 0.999
and 0.997) [20]. The present study systematically assessed different levels of experience, which
makes this study unique. It was found that inexperienced observers measured bigger
liver volumes with both methods and also showed bigger differences between the resected
volume and the real volume. Additionally, they needed longer for the measurement.
However, the differences between the two observer groups were smaller if the measurement
was made with the semiautomated software.
Limitations
Due to the study design, water displacement volumetry for the whole liver could not
be performed. Some studies concerning liver volumetry used explanted livers to measure
the real total liver volume [8]
[17] but in these studies no measurement of resected liver parts was possible. In the
clinical routine it is very important to get a realistic estimate of resected and
remaining liver parts for both hemihepatectomy and living-donor liver transplantation.
In the present study it was also not possible to get a gold standard of the remaining
liver volume after surgery. But this is the part of the liver which is decisive for
patient outcome [21]
[22]. However, it could be shown that the volume of resected liver can be predicted very
well, especially with the semiautomated software, and so the conclusion is acceptable
that the remaining liver volume is predicted equally well.
Another limitation is that resected liver parts were very heterogeneous. Lemke et
al. [2] only used right liver lobes and so the results are more homogenous and unconfounded
by different numbers of segments. The patients in this study were 10 consecutive patients
for partial liver resection and so the study population represents a normal mix of
patients in the clinical routine.
Additionally there is the problem that intraoperative cutting lines are not identical
to the Couinaud segments. Couinaud divides the liver into eight segments along hepatic
veins and the vena porta with straight lines, but the vessels and the corresponding
cutting lines are never straight. Fasel et al. [23] found that up to 51.6 % of the liver area has been attributed to the wrong subsegment
and Fischer et al. [24] stated that the volume of one segment could be overestimated by 24 % or underestimated
by 13 %. In the present study it was a benefit that resected liver segments were already
known when volumetry was performed. In the clinical routine it is an additional problem
that it is only known which segments are planned to be resected and that the real
cutting line during surgery can differ from this. The semiautomated software used
in this study was more accurate in the prediction of resected liver volume, probably
also due to the fact that it was possible to determine the size of every liver segment
as opposed to drawing one straight cutting line through the liver. Especially in cases
with irregular contours of resected segments (e. g. VI – VIII + I), the semiautomated
software should deliver a more accurate volumetry. However, the existing incorrectness
of the Couinaud segments remains a limitation for all preoperative volumetric methods.
Our study population comprised only ten patients, so the semiautomated software should
be tested in a larger population in further studies. Furthermore, the software has
some extra tools which are not tested in the present study, but could be of interest
for future studies, for example the preoperative planning of radiofrequency ablation.
Practical Applications
Both methods for liver volumetry provide an estimated liver volume close to the real
one. The tested semiautomated software is faster, more accurate in predicting resected
liver volume, less dependent on user experience and more reproducible than the manual
method and thus allows a more standardized liver volumetry.
-
This study validates the use of both types of tested software for preoperative evaluation
of patient liver volume (total liver volume and volume of liver parts that are planned
to be resected).
-
The new semiautomated software is considerably faster than the manual method and most
other types of reported software and allows an easy integration of liver volumetry
into routine image evaluation.
-
Additionally, it is less dependent on the user’s experience, which also alleviates
integration into routine imaging.
