Radiology
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online before print November 24, 2004, 10.1148/radiol.2341031801
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2341031801v1
234/1/171    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Hermoye, L.
Right arrow Articles by Van Beers, B. E.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Hermoye, L.
Right arrow Articles by Van Beers, B. E.
(Radiology 2005;234:171-178.)
© RSNA, 2004

Gastrointestinal Imaging

Liver Segmentation in Living Liver Transplant Donors: Comparison of Semiautomatic and Manual Methods1

Laurent Hermoye, MS, Ismael Laamari-Azjal, BS, Zhujiang Cao, MS, Laurence Annet, MD, Jan Lerut, MD, PhD, Benoit M. Dawant, PhD and Bernard E. Van Beers, MD, PhD

1 From the Diagnostic Radiology Unit and Center for Anatomical, Functional and Molecular Imaging Research (L.H., I.L.A., L.A., B.E.V.B.) and Department of Surgery (J.L.), Université Catholique de Louvain, Saint-Luc University Hospital, Ave Hippocrate 10, B-1200 Brussels, Belgium; and Departments of Biomedical Engineering (Z.C.) and Electrical Engineering and Computer Science (B.M.D.), Vanderbilt University, Nashville, Tenn. Received November 7, 2003; revision requested January 28, 2004; final revision received April 13; accepted May 12. Supported in part by grant 3.4578.00 from the Fonds National de la Recherche Scientifique (Belgium) and by NIH grant 4R33CA091352–03. Address correspondence to L.H. (e-mail: hermoye@rdgn.ucl.ac.be).


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
RESULTS
DISCUSSION
 REFERENCES
 
PURPOSE: To compare the accuracy and repeatability of a semiautomatic segmentation algorithm with those of manual segmentation for determining liver volume in living liver transplant donors at magnetic resonance (MR) imaging.

MATERIALS AND METHODS: The institutional review board approved this retrospective study and waived the requirement for informed consent. The semiautomatic segmentation algorithm is based on geometric deformable models and the level-set technique. It entails (a) placing initialization circle(s) on each image section, (b) running the algorithm, (c) inspecting and possibly manually modifying the contours obtained with the segmentation algorithm, and (d) placing lines to separate the liver segments. For 18 living donors (eight men and 10 women; mean age, 34 years; age range, 25–46 years), two observers each performed two semiautomatic and two manual segmentations on contrast material–enhanced T1-weighted MR images. Each measurement was timed. Actual graft weight was measured during surgery. The time needed for manual and that needed for semiautomatic segmentation were compared. Accuracy and repeatability were evaluated with the Bland-Altman method.

RESULTS: Mean interaction time was reduced from 25 minutes with manual segmentation to 5 minutes with semiautomatic segmentation. The mean total time for the semiautomatic process was 7 minutes 20 seconds. Differences between the actual volume and the estimated volume ranged from –223 to +123 mL for manual segmentation and from –214 to +86 mL for semiautomatic segmentation. The 95% limits of agreement for the ratio of actual graft volume to estimated graft volume were 0.686 and 1.601 for semiautomatic segmentation and 0.651 and 1.957 for manual segmentation. Semiautomatic segmentation improved estimation in 15 of 18 cases. Inter- and intraobserver repeatability was higher with semiautomatic segmentation.

CONCLUSION: Use of the semiautomatic segmentation algorithm substantially reduces the time needed for volumetric measurement of liver segments while improving both accuracy and repeatability.

© RSNA, 2004


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
RESULTS
DISCUSSION
 REFERENCES
 
Living donor liver transplantation is increasingly performed as an alternative to cadaveric transplantation (1). The shortage of cadaveric organs and the successful results reported by many centers have fostered the practice. Preoperative screening of the donor candidates is very important (1,2). The quality, size, and vascular and biliary anatomy of the liver are best assessed with magnetic resonance (MR) imaging or computed tomography (CT) (3,4).

In particular, the volume of the potential graft must be measured to ensure sufficient liver function after surgery. In the recipient, the ratio between the graft size and the recipient’s body weight ideally must be higher than 0.8%–1% (5) or the ratio between the graft size and the standard liver volume (which is calculated according to surface area) (6) must be higher than 50% (7). In the donor, a remnant liver volume of 30% of the standard liver volume is usually considered to be the lower limit (8). Lower values are associatedwith higher morbidity and mortality in both the recipient and the donor owing to the small-for-size syndrome. In pediatric patients weighing less than 15 kg, a left lateral segment (Couinaud [9,10] segments II and III) graft is sufficient. In adults, a right lobe (Couinaud segments V–VIII) graft is usually necessary because of volume constraints in the recipient.

Preoperative liver segmentation has proved useful for measuring the graft volume before living donor liver transplantations in previous studies (11–14). In these studies, the liver segments were manually delineated on each image section. The delineated areas were multiplied by the section thickness to obtain volumes and summed to obtain the total volume of the liver segments. This process is tedious and time consuming. To compensate for this problem, automatic segmentation techniques have been proposed (15,16). These methods involve the use of sequences of thresholding, morphologic operations (ie, mathematic operations, such as image dilation, erosion, opening, and closing, that are based on shape), and model deformations. These techniques are complex and require long computation times. Because these algorithms were developed for CT, they are usually not directly usable with MR images, which are increasingly used for the preoperative screening of potential liver donors (3,13). Furthermore, these studies were focused on three-dimensional visualization. The actual organ volume was not used as the standard of reference.

The purpose of our study was to compare the accuracy and repeatability of a semiautomatic segmentation algorithm with those of manual segmentation for determining liver volume in living liver transplant donors at MR imaging.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
DISCUSSION
 REFERENCES
 
Donors
Eighteen subjects were selected as living donors for liver transplantation (eight men [mean age, 34 years; age range, 25–46 years] and 10 women [mean age, 33 years; age range, 27–40 years]). These patients were selected between April 2001 and September 2003. Donors eligible for inclusion in the study were those who passed the preoperative screening tests and whose grafts were weighed after surgery. Eleven donors underwent left lateral segment (segments II and III) resection, two underwent left lobe (segments II–IV) resection, and five underwent right lobe (segments V–VIII) resection. The institutional review board approved this retrospective study and waived the requirement for informed consent.

MR Imaging
Abdominal images were acquired during preoperative screening by using a 1.5-T whole-body MR imaging unit (Gyroscan NT Intera T15; Philips Medical Systems, Best, the Netherlands). To measure the liver volumes, a contrast material–enhanced T1-weighted fast-field-echo sequence was used. Acquisition parameters were as follows: field of view, 400 x 320 mm; acquisition matrix, 304 x 194; reconstruction matrix, 512 x 512; section thickness, 7 mm; section gap, 1–3 mm; repetition time msec/echo time msec, 135/2.9; flip angle, 60°; and acquisition time, 14 seconds per set of images. Twenty sections were acquired to cover the whole liver. The echo time was chosen to create a phase cancellation between fat and water signals. This creates a black outline at the border of the liver in which the voxels contain both liver tissue and peritoneal fat. This effect is beneficial for segmentation. A 0.1-mmol dose of gadodiamide (Omniscan; Amersham Health, Cork, Ireland) per kilogram of body weight was injected at a rate of 2.5 mL/sec with a power injector (Spectris; Medrad, Indianola, Pa). Four sets of images were acquired before administration of the contrast agent and during the arterial, portal venous, and delayed phases. Only the images acquired during the portal venous phase were used for segmentation.

Manual Segmentation
Two independent trained observers (L.H. and I.L.A., who both had 2 years of experience with liver MR imaging) each performed two manual segmentations retrospectively for each patient. To prevent bias, the observers were blinded to the results of their first measurement, the results of the other observer, the results of the other segmentation method, and the actual graft volume. The measurements were performed in random order, and successive measurements in the same patient were performed after a time interval of at least 2 weeks. Manual segmentation was performed at a GE Advantage Windows workstation (GE Medical Systems, Milwaukee, Wis).

The method that was described by Heymsfield et al (17) for measuring the volume of the whole liver and adapted by Kawasaki et al (12) for measuring the segmental volumes was used. On each section, the outlines of the caudate lobe (segment I), the left lateral segment (segments II and III), the left medial segment (segment IV), and the right lobe (segments V–VIII) were manually drawn by using the computer mouse (Fig 1). The anatomic landmarks used to separate the liver segments were the gallbladder and the middle hepatic vein for the right lobe–left medial segment separation; the falciform ligament for the left medial segment–left lateral segment separation; and the inferior vena cava, portal bifurcation, and ligamentum venosum for the delineation of the caudate lobe (18). For sections on which the anatomic landmarks were not visible, the separation lines were mentally interpolated. The volume corresponding to each outline was obtained by multiplying the area of the outline by the section thickness (including the gap). The total volumes of the liver segments were obtained by summing the volume on each section. For each measurement, the total segmentation time was recorded.



View larger version (132K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1a. (a) Manual and (b) semiautomatic segmentation of caudate lobe (segment I), left lateral segment (segments II and III), left medial segment (segment IV), and right lobe (segments V-VIII) on representative transverse contrast-enhanced T1-weighted fast-field-echo MR images (135/2.9; flip angle, 60°).

 


View larger version (118K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1b. (a) Manual and (b) semiautomatic segmentation of caudate lobe (segment I), left lateral segment (segments II and III), left medial segment (segment IV), and right lobe (segments V-VIII) on representative transverse contrast-enhanced T1-weighted fast-field-echo MR images (135/2.9; flip angle, 60°).

 
Semiautomatic Segmentation
The same two observers each retrospectively performed two semiautomatic segmentations for each patient. The same measures were taken to prevent bias. Our semiautomatic liver segmentation software was developed at a Windows 2000 workstation with the Interactive Data Language programming language, version 5.5 (Research Systems, Boulder, Colo). A user-friendly graphical user interface allows easy interaction. The algorithm used in this study is an extension of the algorithm proposed by Pan and Dawant (19) for segmentation of the liver in CT images. Briefly, this algorithm requires placing one or several small circles within the structure of interest. These circles are then deformed automatically until they reach the structure boundary. The technique used for deforming the circles falls within the category of geometric deformable models (20–22) and is implemented in a level-set framework (23,24).

Specific features of the algorithm used herein include a speed function called the accumulative speed function (19), which permits stopping the progression of the contours at weak boundaries, automatic estimation of the parameters required by this algorithm on the basis of the initial circles, and a modification of the algorithm initially proposed by Pan and Dawant to prevent early contour stopping caused by the presence of the hepatic vessels that appear as very bright structures in our images. The overall segmentation process involves the following steps: (a) placing initialization circle(s) on each section, (b) running the algorithm, and (c) inspecting and possibly manually modifying the contours obtained with the segmentation algorithm.

With the segmentation algorithm, two lines must be drawn to separate the liver into segments. As in the manual segmentation process, the liver is separated into the caudate lobe, left lateral segment, left medial segment, and right lobe (Fig 1). Separation is performed by drawing lines on a few sections on which the landmarks are visible. These lines are then automatically interpolated to create surfaces that separate the liver into segments. The caudate lobe must be manually delineated. Each measurement was timed. The initialization time (ie, the time required to place the initialization circles), the computing time (ie, the time required to automatically delineate the liver contours), and the time for manual corrections and delineation of the liver segments were recorded separately. The total time was the sum of these three times. The interaction time was the sum of the initialization time and the time for manual corrections and delineation of the liver segments.

Surgical Procedure
The recipients received a left lateral segment (segments II and III), left lobe (segments II–IV), or right lobe (segments V–VIII) graft (25). Intraoperative ultrasonography was used in the donors to delineate the intrahepatic vascular anatomy. The arterial, venous, and biliary elements of the corresponding graft were isolated. Parenchymal transection was performed without inflow occlusion. The liver graft was flushed successively with Hartmann (4°C) and University of Wisconsin solutions through a portal vein cannula. The corresponding hepatic artery was then ligated, and the corresponding hepatic vein was transected and sutured progressively. The graft was weighed by using a calibrated scale after the flushing and the preparation work. Graft weight was converted into graft volume by using an assumed unit density for the liver (26).

Effect of Section Thickness
The influence of the section thickness on the volumes measured with the semiautomatic segmentation algorithm was assessed in five patients with normal livers. In each patient, the contrast-enhanced T1-weighted fast-field-echo sequence used for preoperative screening of the living donors was performed three times with increasing section thicknesses (5, 7, and 10 mm). The section gap (1 mm) was not modified. For the images acquired with each sequence, one observer (L.H.) performed two semiautomatic segmentations as described above. The estimated volumes of the caudate lobe (segment I), the left lateral segment (segments II and III), the left medial segment (segment IV), the right lobe (segments V–VIII), and the total liver, as averaged across the five patients, are reported for the three section thicknesses. The influences of section thickness on the number of sections necessary to cover the whole liver and on the interaction and total segmentation times were also studied.

Statistical Analysis
The total time for manual segmentation was compared with the total and interaction times for semiautomatic segmentation. For the semiautomatic segmentation method, the percentages of the total time devoted to initialization, computing, and manual corrections and delineation of the liver segments were calculated.

For accuracy comparisons, the mean graft volumes ± standard deviations measured during surgery and the mean volumes (ie, the mean of the four observations [two observations by each of two observers]) estimated with the manual and semiautomatic segmentations were averaged across all of the donors. In addition, for each donor, the difference between the actual graft volume (measured during surgery) and the estimated volume (the mean of the four observations) was calculated separately for the manual and semiautomatic segmentations. These differences were plotted against the averages of the actual and the estimated volumes. Scatterplots were drawn to illustrate the relationships between the actual and estimated graft volumes. The identity line was indicated on these plots. The number of cases in which use of the semiautomatic segmentation algorithm improved the accuracy of the estimation was counted.

The statistical analysis was performed with the method described by Bland and Altman (27,28). The 95% limits of agreement were calculated separately for the actual volume versus the volume determined with manual segmentation and the actual volume versus the volume determined with semiautomatic segmentation. Because the measurement error was proportional to the mean, a logarithmic transformation (natural logarithm) was used (29). Independence after the transformation was checked by using a Bland and Altman plot (27,28). Normality of the differences was checked with the Kolmogorov-Smirnov test. Because for each segmentation method the mean of four measurements was used, the variance of the differences was corrected for repeated measurements (28). The within-subject variance was calculated with one-way analysis of variance. The 95% limits of agreement were calculated as the mean difference ± 1.96 (adjusted standard deviation of the differences). The limits of agreement expressed in a logarithmic scale were transformed back into being expressed in a natural scale by using an exponential function. Because the calculations were performed with the log-transformed quantities, these limits of agreement must be interpreted in terms of ratios.

The Bland and Altman repeatability coefficient (28,30) was used as a measurement of inter- and intraobserver agreement for the manual and semiautomatic segmentations. The calculations were performed separately for the caudate lobe (segment I), the left lateral segment (segments II and III), the left medial segment (segment IV), the right lobe (segments V–VIII), and the whole liver. Because for each of these tests the range of values was narrow, logarithmic transformation was not necessary. For the measurement of interobserver agreement, only the first measurement of each observer was taken into account (31). Intraobserver agreement was measured for both observers. The repeatability coefficient was calculated as 2.77 times the within-subject standard deviation. By definition, the measurement error is smaller than the repeatability coefficient for 95% of the observations. Therefore, a smaller repeatability coefficient indicates better repeatability of the method.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
DISCUSSION
 REFERENCES
 
Time Measurement
The time measurements (expressed as means ± standard deviations), as averaged across the four observations, are reported in Table 1. The interaction time was reduced from 25 minutes for manual segmentation to 5 minutes for semiautomatic segmentation. For semiautomatic segmentation, interaction time represents 67% of the total time required and is divided into the time required for initialization of the contours (21% of the total time) and the time required for corrections and placing of the separation lines (46% of the total time). Computing time represents 33% of the total time.


View this table:
[in this window]
[in a new window]

  		TABLE 1. Timing Involved in Segmentation Procedures

 
Accuracy Measurements
The mean graft volumes ± standard deviations that were measured during surgery and the mean graft volumes (ie, the mean of the four observations) that were estimated with the manual and semiautomatic segmentation methods are summarized in Table 2. The differences between the actual volume and the estimated volume ranged from –223 to +123 mL for manual segmentation and from –214 to +86 mL for semiautomatic segmentation (Fig 2). The volumes of larger grafts (whether involving the left or the right lobe) were always overestimated. It can be observed on the scatterplots (Fig 3) that the experimental points are closer to the identity line for the semiautomatic than for the manual segmentation values. Semiautomatic segmentation improved graft volume estimation in 15 of the 18 cases.


View this table:
[in this window]
[in a new window]

  		TABLE 2. Actual and Estimated Graft Volumes

 


View larger version (16K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2a. Plots of differences between actual and (a) manually and (b) semiautomatically estimated volumes against their averages. The degree of dispersion around the horizontal axis is smaller for the semiautomatic segmentation values. An increase in measurement error with increasing volume can be observed for both methods.

 


View larger version (16K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2b. Plots of differences between actual and (a) manually and (b) semiautomatically estimated volumes against their averages. The degree of dispersion around the horizontal axis is smaller for the semiautomatic segmentation values. An increase in measurement error with increasing volume can be observed for both methods.

 


View larger version (25K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3a. Scatterplots of actual versus (a) manually and (b) semiautomatically determined volumes. The experimental points are closer to the dashed identity line for semiautomatic than for manual segmentation values.

 


View larger version (25K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3b. Scatterplots of actual versus (a) manually and (b) semiautomatically determined volumes. The experimental points are closer to the dashed identity line for semiautomatic than for manual segmentation values.

 
After logarithmic transformation, the dependence of the measurement errors on the mean was removed for both comparisons. Applying the Kolmogorov-Smirnov test to the logarithmic differences did not lead to the rejection of the normality hypotheses (P = .786 and P = .828 for actual volume vs manual and semiautomatic segmentations, respectively). The mean, the variance, the standard deviation of the differences, the within-subject variance, the adjusted variance, and the adjusted standard deviation of the accuracy measurements are summarized in Table 3 for both tests. The 95% limits of agreement are also indicated in logarithmic and natural scales.


View this table:
[in this window]
[in a new window]

  		TABLE 3. Accuracy Measurements

 
The mean difference was greater for manual segmentation than for semiautomatic segmentation. The smaller within-subject variance obtained with semiautomatic segmentation was a first indicator of the better repeatability of this method. The ratio of actual graft volume to estimated graft volume lay between 0.651 and 1.957 for manual segmentation and between 0.686 and 1.601 for semiautomatic segmentation. Because the 95% limits of agreement were closer to 1 for the semiautomatic than for the manual method, the accuracy of the former method was higher.

Repeatability Measurements
The repeatability coefficients for both the manual and the semiautomatic segmentation methods are reported in Table 4. Lower repeatability coefficients (ie, better repeatability) were obtained with the semiautomatic segmentation method for all segments except the caudate lobe (segment I).


View this table:
[in this window]
[in a new window]

  		TABLE 4. Repeatability Coefficients

 
Effect of Section Thickness
The estimated volumes of the caudate lobe (segment I), left lateral segment (segments II and III), left medial segment (segment IV), right lobe (segments V–VIII), and whole liver; the number of sections; and the total and interaction times are reported in Table 5 as means and standard deviations averaged across the five patients for the three section thicknesses. A decrease in the mean estimated volume with increasing section thickness was observed for the left lateral segment, the right lobe, and the whole liver. When the section thickness was doubled (from 5 to 10 mm), the mean estimated volumes of these areas decreased by 8%, 4%, and 5%, respectively. No specific trend was observed for the caudate lobe and the left medial segment. Because with the use of thinner sections it was necessary to acquire more sections to cover the whole liver, the interaction and total times were higher.


View this table:
[in this window]
[in a new window]

  		TABLE 5. Influence of Section Thickness

 

   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
DISCUSSION
REFERENCES
 
In liver transplantation, the graft size cannot be predicted according to the donor’s body weight (12). Therefore, determination of hepatic volumes with cross-sectional imaging is mandatory before living related liver transplantation (1,2,12). We have developed a semiautomatic segmentation method that has been optimized for the determination of liver volumes at MR imaging because this method is being increasingly used in living related liver transplantation to reveal the arterial, portal venous, hepatic venous, and biliary anatomy (3,13). The use of T1-weighted opposed-phase MR images helped define the boundaries of the liver. The determination of the limits of the liver segments was based on the simple surfaces defined by the Couinaud nomenclature system (9,10). Use of the semiautomatic segmentation algorithm greatly decreased the time needed for the preoperative determination of liver volumes. The reduction of the mean user interaction time from 25 minutes to less than 5 minutes made liver segmentation at MR imaging easier and much faster. Both accuracy and repeatability were improved.

The main advantage of the semiautomated method is the time savings it confers. Initialization time might be further reduced by using three-dimensional segmentation. In this case, initialization on one or two sections would be sufficient. Automatic initialization could even be considered. However, three-dimensional segmentation would increase the computation time. Furthermore, its accuracy would have to be evaluated. Manual corrections and placing of separation lines are the most time-consuming steps. If the initialization circles are judiciously placed, manual corrections are only necessary at the borders, at which the contrast with the surrounding organs is very weak. This occurs especially in lean patients because in other patients, phase cancellation between the liver and the surrounding fat creates a black line surrounding the liver.

The improvement in accuracy can be explained by the more precise delineation of the liver contours. The accuracy of manual segmentation is indeed limited by the observer’s dexterity. If the automatic segmentation algorithm fails, manual corrections can quickly remove the mistakes. The positioning of lines on the anatomic landmarks that separate the segments and the automatic interpolation of these lines on the sections on which the landmarks are not visible are also favorable factors. Furthermore, the apparent border of the liver depends on the intensity window center and width. The manual adjustment of image contrast can thus influence the volume measured during manual segmentation (32), whereas automatic segmentation is independent of this setting. Nevertheless, even with semiautomated segmentation, a factor in the range of 0.686–1.601 can still separate the estimated volume from the actual graft size. In adult grafts (whether from the right or the left lobe), an overestimation and an underestimation of graft size would lead to a higher risk of small-for-size syndrome in the recipient and in the donor, respectively.

Several reasons can explain the discrepancies observed between the estimated and the real volumes. The surgical context may impose the necessity of using slightly different anatomic landmarks to separate the liver segments. Furthermore, during the preparation of the graft, the blood vessels are flushed. Therefore, the blood vessels included in the segmented liver could lead to an overestimation of the graft size, on the order of 29% (33). Because this effect is more important for right- or left-lobe grafts, this may explain the systematic overestimation observed for these grafts. Partial volume effects related to the stomach and the heart are often present in upper sections (12). On these sections, the delineation of the liver must be based on both the image contrast and the contours on the adjacent sections.

Section thickness is also a limiting factor. With 7-mm sections and 3-mm gaps, the liver border is averaged over 10 mm. As shown in the results of our study, the left lateral segment, the right lobe, and the whole liver are the areas whose measurements are most affected by the section thickness. These liver parts have some oblique interfaces (relative to the imaging plane) with the surrounding organs (Fig 1) and are therefore sensitive to partial volume effects. In contrast, the borders of the caudate lobe and the left medial segment are more perpendicular to the imaging plane and are less sensitive to partial volume effects. Similar results were reported previously (34). It was shown that estimations of the volume of the bladder, whose shape varies along the z-axis, were higher with thinner sections, whereas the volume of the rectum, a cylindric organ that is quasiperpendicular to the imaging plane, was insensitive to the thickness of transverse sections.

In the present study, the actual and estimated graft volumes were similar to those reported in previous studies (12,35). The differences between the actual and the estimated graft volumes were in the normal range (11). Because most previous studies (11,12,35) involved the use of correlation coefficients and regression analysis to quantify the relationship between actual and estimated graft volumes, comparison of the 95% limits of agreement is difficult. Because the use of correlation coefficients and regression analysis is debated in the context of methods for comparing studies (27), their use was avoided. The main drawback of correlation coefficients and regression analysis relates to the fact that they measure the strength of the linear relationship between two variables, not the agreement between them. Furthermore, they depend on the scale and range of the measurements. Data that seem to be in poor agreement can produce quite high correlations (27). The statistical methods used in this study are similar to those used by Bakker et al (31) for renal volume measurements.

The use of automatic processes always increases repeatability (36). As expected, use of the semiautomatic segmentation method led to higher intra- and interobserver repeatability (except for the caudate lobe, for which the segmentation was manual in both methods). Manual determination of the liver contours, mental interpolation of the separations between the segments, and dependence on the display window settings (32) can explain the lower repeatability achieved with manual segmentation. However, even with semiautomatic segmentation, several variability sources exist. The manual placement of the initialization circles slightly modifies the final segmentation. The best results are obtained when two or three initialization circles are placed in representative areas on each section to take variations in the hepatic signal intensity into account. Manual corrections are left to the observer’s own judgment. Interobserver variability can result, especially in sections in which partial volume effects are present. The positioning of the separation lines between the liver segments can also introduce variability in patients in whom the anatomic landmarks are unclear.

The study had several limitations. Only 18 donors were included. This comparison could be performed with a larger number of donors within the framework of a multicenter study. The effects of the inclusion of blood vessels in the segmented liver and those of partial volume effects and section thickness have already been discussed. Moreover, in this study, the actual graft volume was inferred from its weight (11,12,35) while assuming a unit density for the liver (26). Water displacement (37) would allow direct measurement of the graft volume but would be cumbersome in clinical practice.

Finally, the accuracy of determining the segmental and subsegmental boundaries with straight lines has been questioned (18,38). The true anatomic limits of the liver subsegments are determined by the portal anatomy, which shows large variations (38). To take this variability into account, the use of liver segmentation algorithms based on the portal anatomy has been proposed (15,39). However, these methods are complex and are not widely used. In addition, most previous analyses of discrepancies between radiologic determinations of the segmental anatomy and the actual anatomy of liver segments and subsegments have concerned the limits between the subsegments of the right lobe (38,40,41). In contrast, the limits used in the present study—namely, the limits between the left lateral segment and the left medial segment and those between the left and the right lobe—conform fairly to vertical planes (42).

In conclusion, in this validation study performed with living donors for liver transplantation, the use of semiautomatic liver segmentation saved a substantial amount of time while improving both accuracy and repeatability. Other potential applications of the technique include the determination of segmental volumes before partial hepatectomy for tumors (43) and the monitoring of liver regeneration (44).


   FOOTNOTES
 
Authors stated no financial relationship to disclose.

Author contributions: Guarantors of integrity of entire study, L.H., B.E.V.B.; study concepts, L.H., Z.C., J.L., B.M.D., B.E.V.B.; study design, L.H., I.L.A., B.M.D., B.E.V.B.; literature research, L.H., I.L.A., J.L., B.E.V.B.; clinical studies, L.H., I.L.A.; data acquisition, L.H., I.L.A., Z.C., J.L., L.A.; data analysis/interpretation, L.H., I.L.A., J.L., L.A., B.E.V.B.; statistical analysis, L.H.; manuscript preparation and editing, L.H.; manuscript definition of intellectual content, revision/review, and final version approval, all authors


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
DISCUSSION
 REFERENCES
 

  1. Broering DC, Sterneck M, Rogiers X. Living donor liver transplantation. J Hepatol 2003; 38(suppl 1):S119-S135.
  2. Chen YS, Cheng YF, de Villa VH, et al. Evaluation of living liver donors. Transplantation 2003; 75:S16-S19.[CrossRef][Medline]
  3. Cheng YF, Chen CL, Huang TL, et al. Single imaging modality evaluation of living donors in liver transplantation: magnetic resonance imaging. Transplantation 2001; 72:1527-1533.[CrossRef][Medline]
  4. Schroeder T, Nadalin S, Stattaus J, Debatin JF, Malago M, Ruehm SG. Potential living liver donors: evaluation with an all-in-one protocol with multi-detector row CT. Radiology 2002; 224:586-591.[Abstract/Free Full Text]
  5. Kiuchi T, Kasahara M, Uryuhara K, et al. Impact of graft size mismatching on graft prognosis in liver transplantation from living donors. Transplantation 1999; 67:321-327.[Medline]
  6. Urata K, Kawasaki S, Matsunami H, et al. Calculation of child and adult standard liver volume for liver transplantation. Hepatology 1995; 21:1317-1321.[CrossRef][Medline]
  7. Kawasaki S, Makuuchi M, Matsunami H, et al. Living related liver transplantation in adults. Ann Surg 1998; 227:269-274.[CrossRef][Medline]
  8. Fan ST, Lo CM, Liu CL, Yong BH, Chan JK, Ng IO. Safety of donors in live donor liver transplantation using right lobe grafts. Arch Surg 2000; 135:336-340.[Abstract/Free Full Text]
  9. Couinaud C. Liver anatomy: portal (and suprahepatic) or biliary segmentation. Dig Surg 1999; 16:459-467.[CrossRef][Medline]
  10. Soyer P. Segmental anatomy of the liver: utility of a nomenclature accepted worldwide. AJR Am J Roentgenol 1993; 161:572-573.[Abstract/Free Full Text]
  11. Sakamoto S, Uemoto S, Uryuhara K, et al. Graft size assessment and analysis of donors for living donor liver transplantation using right lobe. Transplantation 2001; 71:1407-1413.[CrossRef][Medline]
  12. Kawasaki S, Makuuchi M, Matsunami H, et al. Preoperative measurement of segmental liver volume of donors for living related liver transplantation. Hepatology 1993; 18:1115-1120.[CrossRef][Medline]
  13. Fulcher AS, Szucs RA, Bassignani MJ, Marcos A. Right lobe living donor liver transplantation: preoperative evaluation of the donor with MR imaging. AJR Am J Roentgenol 2001; 176:1483-1491.[Abstract/Free Full Text]
  14. Vauthey JN, Abdalla EK, Doherty DA, et al. Body surface area and body weight predict total liver volume in Western adults. Liver Transpl 2002; 8:233-240.[CrossRef][Medline]
  15. Soler L, Delingette H, Malandain G, et al. Fully automatic anatomical, pathological, and functional segmentation from CT scans for hepatic surgery. Comput Aided Surg 2001; 6:131-142.[CrossRef][Medline]
  16. Gao L, Heath DG, Kuszyk BS, Fishman EK. Automatic liver segmentation technique for three-dimensional visualization of CT data. Radiology 1996; 201:359-364.[Abstract/Free Full Text]
  17. Heymsfield SB, Fulenwider T, Nordlinger B, Barlow R, Sones P, Kutner M. Accurate measurement of liver, kidney, and spleen volume and mass by computerized axial tomography. Ann Intern Med 1979; 90:185-187.
  18. Strunk H, Stuckmann G, Textor J, Willinek W. Limitations and pitfalls of Couinaud’s segmentation of the liver in transaxial imaging. Eur Radiol 2003; 13:2472-2482.[CrossRef][Medline]
  19. Pan S, Dawant BM. Automatic 3D segmentation of the liver from abdominal CT images: a level set approach. Proc SPIE Med Imaging 2001; 4322:128-138.
  20. Caselles V, Catte F, Coll T, Dibos F. A geometric model for active contours. Numerische Mathematik 1993; 66:1-31.
  21. Malladi R, Sethian JA, Vemuri BC. Evolutionary fronts for topology independent shape modeling and recovery. Proceedings of the 3rd European Conference on Computer Vision 1994; 3-13.
  22. Malladi R, Sethian JA, Vemuri BC. Shape modeling with front propagation: a level set approach. IEEE Trans Pat Anal Mach Intel 1995; 17:158-175.[CrossRef]
  23. Osher S, Sethian JA. Fronts propagating with curvature-dependent speed: algorithms based on Hamilton-Jacobi formulations. J Comput Phys 1988; 79:12-49.[CrossRef]
  24. Sethian JA. Numerical algorithms for propagating interfaces: Hamilton-Jacobi equations and conservation laws. J Dif Geom 1990; 31:131-161.
  25. Lerut JP, Ciccarelli O, Roggen FM, et al. Adult-to-adult living related liver transplantation: initial experience. Acta Gastroenterol Belg 2001; 64:9-14.[Medline]
  26. Van Thiel DH, Hagler NG, Schade RR, et al. In vivo hepatic volume determination using sonography and computed tomography: validation and a comparison of the two techniques. Gastroenterology 1985; 88:1812-1817.[Medline]
  27. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 1:307-310.[CrossRef][Medline]
  28. Bland JM, Altman DG. Measuring agreement in method comparison studies. Stat Methods Med Res 1999; 8:135-160.[Abstract/Free Full Text]
  29. Bland JM, Altman DG. Measurement error proportional to the mean. BMJ 1996; 313:106.[Free Full Text]
  30. Bland JM, Altman DG. Measurement error. BMJ 1996; 313:744.[Free Full Text]
  31. Bakker J, Olree M, Kaatee R, et al. Renal volume measurements: accuracy and repeatability of US compared with that of MR imaging. Radiology 1999; 211:623-628.[Abstract/Free Full Text]
  32. Koehler PR, Anderson RE, Baxter B. The effect of computed tomography viewer controls on anatomical measurements. Radiology 1979; 130:189-194.[Medline]
  33. Hwang S, Lee SG, Kim KH, et al. Correlation of blood-free graft weight and volumetric graft volume by an analysis of blood content in living donor liver grafts. Transplant Proc 2002; 34:3293-3294.[CrossRef][Medline]
  34. Berthelet E, Liu M, Truong P, et al. CT slice index and thickness: impact on organ contouring in radiation treatment planning for prostate cancer. J Appl Clin Med Phys 2003; 4:365-373.[CrossRef][Medline]
  35. Leelaudomlipi S, Sugawara Y, Kaneko J, Matsui Y, Ohkubo T, Makuuchi M. Volumetric analysis of liver segments in 155 living donors. Liver Transpl 2002; 8:612-614.[CrossRef][Medline]
  36. Bellon E, Feron M, Maes F, et al. Evaluation of manual vs semi-automated delineation of liver lesions on CT images. Eur Radiol 1997; 7:432-438.[CrossRef][Medline]
  37. Luft AR, Skalej M, Welte D, Kolb R, Klose U. Reliability and exactness of MRI-based volumetry: a phantom study. J Magn Reson Imaging 1996; 6:700-704.[Medline]
  38. Fasel JH, Selle D, Evertsz CJ, Terrier F, Peitgen HO, Gailloud P. Segmental anatomy of the liver: poor correlation with CT. Radiology 1998; 206:151-156.[Abstract/Free Full Text]
  39. Selle D, Preim B, Schenk A, Peitgen HO. Analysis of vasculature for liver surgical planning. IEEE Trans Med Imaging 2002; 21:1344-1357.[CrossRef][Medline]
  40. van Leeuwen MS, Noordzij J, Hennipman A, Feldberg MA. Planning of liver surgery using three dimensional imaging techniques. Eur J Cancer 1995; 31A:1212-1215.
  41. Ohashi I, Ina H, Okada Y, et al. Segmental anatomy of the liver under the right diaphragmatic dome: evaluation with axial CT. Radiology 1996; 200:779-783.[Abstract/Free Full Text]
  42. Fischer L, Cardenas C, Thorn M, et al. Limits of Couinaud’s liver segment classification: a quantitative computer-based three-dimensional analysis. J Comput Assist Tomogr 2002; 26:962-967.[CrossRef][Medline]
  43. Okamoto E, Kyo A, Yamanaka N, Tanaka N, Kuwata K. Prediction of the safe limits of hepatectomy by combined volumetric and functional measurements in patients with impaired hepatic function. Surgery 1984; 95:586-592.[Medline]
  44. Maetani Y, Itoh K, Egawa H, et al. Factors influencing liver regeneration following living-donor liver transplantation of the right hepatic lobe. Transplantation 2003; 75:97-102.[CrossRef][Medline]



This article has been cited by other articles:


Home page
 RadiologyHome page
Y. Nakayama, Q. Li, S. Katsuragawa, R. Ikeda, Y. Hiai, K. Awai, S. Kusunoki, Y. Yamashita, H. Okajima, Y. Inomata, et al.
Automated Hepatic Volumetry for Living Related Liver Transplantation At Multisection CT
Radiology, September 1, 2006; 240(3): 743 - 748.
[Abstract] [Full Text] [PDF]

Home page
 RadiologyHome page
A.-J. Lemke, M. J. Brinkmann, T. Schott, S. M. Niehues, U. Settmacher, P. Neuhaus, and R. Felix
Living Donor Right Liver Lobes: Preoperative CT Volumetric Measurement for Calculation of Intraoperative Weight and Volume
Radiology, September 1, 2006; 240(3): 736 - 742.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2341031801v1
234/1/171    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Hermoye, L.
Right arrow Articles by Van Beers, B. E.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Hermoye, L.
Right arrow Articles by Van Beers, B. E.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
RADIOLOGY RADIOGRAPHICS RSNA JOURNALS ONLINE