Eur Radiol (2004) 14:326–333 DOI 10.1007/s00330-003-2161-8

Bernd B. Frericks Franco C. Caldarone Björn Nashan Dagmar Högemann Savellano Georg Stamm Timm D. Kirchhoff Hoen-Oh Shin Andrea Schenk Dirk Selle Wolf Spindler Jürgen Klempnauer Heinz-Otto Peitgen Michael Galanski Received: 19 November 2002 Revised: 10 June 2003 Accepted: 24 October 2003 Published online: 10 December 2003 © Springer-Verlag 2003 B. B. Frericks (✉) · F. C. Caldarone D. H. Savellano · G. Stamm · T. D. Kirchhoff · H.-O. Shin · M. Galanski Medizinische Hochschule Hannover, Diagnostische Radiologie, Hannover, Germany e-mail: [email protected] Tel.: +49-30-84453041 Fax: +49-30-84454474 B. Nashan · J. Klempnauer Medizinische Hochschule Hannover, Viszeral und Transplantationschirurgie, Hannover, Germany A. Schenk · D. Selle · W. Spindler H.-O. Peitgen Centrum für Medizinische Diagnosesysteme und Visualisierung, Bremen, Germany Present address: B. B. Frericks, Klinik und Poliklinik für Radiologie und Nuklearmedizin, Universitätsklinikum Benjamin Franklin, Freie Universität Berlin, Hindenburgdamm 30, 12200 Berlin, Germany

H E PAT O B I L I A R Y – PA N C R E A S

3D CT modeling of hepatic vessel architecture and volume calculation in living donated liver transplantation

Abstract The aim of this study was to evaluate a software tool for noninvasive preoperative volumetric assessment of potential donors in living donated liver transplantation (LDLT). Biphasic helical CT was performed in 56 potential donors. Data sets were post-processed using a non-commercial software tool for segmentation, volumetric analysis and visualisation of liver segments. Semi-automatic definition of liver margins allowed the segmentation of parenchyma. Hepatic vessels were delineated using a region-growing algorithm with automatically determined thresholds. Volumes and shapes of liver segments were calculated automatically based on individual portal-venous branches. Results were visualised three-dimensionally and statistically compared with conventional volumetry and the intraoperative findings in 27 transplanted cases. Image processing was easy to perform within 23 min. Of the 56 potential donors, 27 were excluded from LDLT because of inappropriate

Introduction Living donated liver transplantation (LDLT) was developed to overcome the shortage of suitably sized organs for children and adults with end-stage liver disease. Since impaired liver function after resection and transplantation is caused by insufficient liver volume [1], a reliable volumetric assessment of the hepatic segments

liver parenchyma or vascular architecture. Two recipients were not transplanted due to poor clinical conditions. In the 27 transplanted cases, preoperatively visualised vessels were confirmed, and only one undetected accessory hepatic vein was revealed. Calculated graft volumes were 1110±180 ml for right lobes, 820 ml for the left lobe and 270±30 ml for segments II+III. The calculated volumes and intraoperatively measured graft volumes correlated significantly. No significant differences between the presented automatic volumetry and the conventional volumetry were observed. A novel image processing technique was evaluated which allows a semiautomatic volume calculation and 3D visualisation of the different liver segments. Keywords CT · Liver transplantation · Volumetry · Vessel segmentation

of potential living donors is one of the key factors in the preoperative donor evaluation. Recently, preoperative liver volumetry based on CT volume data sets resulted in significantly improved clinical outcome [2, 3, 4]; however, these procedures require an arbitrary definition of the potential resection line along the middle hepatic vein and are therefore relatively operator dependent. Although several refinements in semi-automatic liver volu-

327

metry have been introduced, leading to individualised computer-generated resection protocols [5], there is no report concerning automatic liver segmentation and volumetry on the basis of the intrahepatic portal venous blood supply. In the present study, an evaluation protocol was assessed in terms of reliability and accuracy, by employing a new mathematical model which allows for a semi-automatic volumetric analysis and visualisation of the different liver segments based on the ramification pattern of the portal venous branches.

Methods Donors and recipients Within 3 years, 56 consecutive potential donors for LDLT underwent CT examinations as part of the routine preoperative evaluation protocol. The potential donors were 28 men and 28 women (age range 19–59 years, mean age 38 years). In accordance with the requirements of our institutional review board, all potential donors were informed that the CT examination was performed for delineation of the hepatic vessel anatomy, and volumetric assessment of the different liver parts and written informed consent was obtained prior to CT. The recipient group consisted of 53 patients [24 children, 2 adolescents (11 and 14 years) and 27 adults] suffering from hepatic failure. Underlying diseases were biliary atresia (n=20), Allagille’s syndrome (n=1), Alpha-1-antitrypsine deficiency (n=1), acute liver failure of unknown aetiology (n=2) in the children recipient group, cryptogenic cirrhosis (n=9), cirrhosis due to hepatitis B and C (n=8), hepatocellular carcinoma (on basis of hepatitis B and C cirrhosis, n=5; on basis of hepatitis B and D cirrhosis, n=1; on basis of tyrosinaemia, n=1), and primary sclerosing cholangitis, amyloidosis, autoimmune hepatitis, cystic liver degeneration and neuroendocrine metastases (n=1 each) in the adolescent and adult recipient group. In 27 potential donors, findings of CT imaging and data processing allowed partial liver donation, and transplantations were carried out within 41±37 days (range 1–159 days) after the respective CT scan. These donors were included in the analysis of volumetric liver assessment since a direct comparison of the resected liver graft volume and the preoperatively calculated volume was possible. Imaging technique Helical CT scans were performed from November 1998 to October 2000 using a Somatom Plus 4 single-slice scanner (Siemens, Erlangen, Germany) and from October 2000 to September 2001 using a Lightspeed QX/i 4-detector-row CT scanner (GE Medical Systems, Milwaukee, Wis.). All scans were obtained during inspiration and covered the whole liver parenchyma. The first CT scans were carried out, either non-enhanced (n=51) or enhanced (n=5), after intravenous administration of 50 ml of a cholangiographic contrast agent containing 108 mg iodine/ml (Biliscopin, Schering, Berlin, Germany). Afterwards, 150 ml of a nonionic contrast agent containing 300 mg iodine/ml (623 mg Iopromid/ml; Ultravist, Schering, Berlin, Germany) were administered intravenously at a rate of 4 ml/s. Contrast-enhanced scanning in the arterial phase was started automatically 5 s after bolus triggering in the abdominal aorta. A second scan was performed during the venous phase starting 70 s after initiating the contrast agent injection for exact delineation of the portal and hepatic veins. Scan parameters were 3-mm collimation, 140 kV and 206 mAs (rotation time 1 s) for the Somatom Plus 4 scanner; and 2.5-mm collimation, 140 kV and 147 mAs (rotation time 0.7 s) for the Lightspeed QX/I scanner.

Using an imaging matrix of 512¥512 pixels and an individually adjusted field of view, the pixel size ranged from 0.63–0.88 mm2. Both the arterial and the venous phase were reconstructed at an interval of 2 mm to obtain the same number of slices necessary for successful matching. Data processing of liver parenchyma and hepatic vessels The uncompressed data were transferred to a workstation (Octane, Silicon Graphics) and post-processed using a non-commercial software-tool (HepaVision, MeVis, Center for Medical Diagnostic Systems and Visualization, Bremen, Germany) [6]. This research software was developed in close cooperation between radiologists and information technologists and is a dedicated system for segmentation, volumetric analysis and visualisation of anatomical and pathological structures of the liver. Particular tools allow the delineation of liver segments depending on their individual portal venous supply. Data processing was carried out by three radiologists equally experienced in image analysis using this software tool (B.B.F., F.C.C., D.H.S.). Two of them had had previous training in liver surgery (B.B.F., D.H.S.). Each data set was processed once by one of the three radiologists (B.B.F., F.C.C., D.H.S.). Only in problematic cases was data processing repeated by one of the other two radiologists independently, and consensus was achieved. Complete evaluation of the liver of a potential donor required four steps: (a) segmentation of liver parenchyma; (b) segmentation of hepatic vessels; (c) separation and analysis of hepatic arteries, portal veins and hepatic veins; and (d) determination of the individual vascular territories. The time required for each of the four steps was recorded. Segmentation of liver parenchyma was achieved employing an algorithm with real-time computation of optimal boundary paths between a few user-defined contour points: in real-time, a connection between the last point and the current position of the cursor was calculated considering intensity and edge information of the liver. This user-steered segmentation was performed on every fifth CT image obtained during the venous phase, and all intermediate contours were calculated automatically by shape-based interpolation [7, 8]. All interpolated contours for which interactive corrections became necessary in order to separate neighbouring extrahepatic structures of equal density, such as the diaphragm, intercostal muscles and the stomach wall, were recorded. All large vessels, including the inferior vena cava and extrahepatic portal vein, and major fissures, were carefully excluded. After the application of a filter for noise reduction and background compensation, the hepatic vessels were segmented using a region-growing algorithm. Starting from a seed point, a growing region was included in all 3D voxels (image points) with intensities between the maximum and a threshold. The program computed results for decreasing thresholds and automatically suggested an effective threshold. Subsequent interactive adjustment of this value was possible, and recorded when necessary, for optimised visualisation of higher-order vascular branches. The resulting segmented vascular structures underwent further processing by a “skeletonisation” algorithm that computes centre lines, radii and lengths of all vessel branches. Based on this information, the morphology, ramification pattern and hierarchy of the vascular systems were automatically determined [8] and used for the separation and classification of arteries, portal veins and hepatic veins [9]. As described previously, respective analysis of the portal venous system allowed partitioning of the liver parenchyma into its vascular territories with an automatic calculation of the respective absolute and relative volumes [10]. To explore the extracted anatomical structures, a choice of visualisation techniques, including direct volume rendering and shaded-surface displays, were provided. Viewing directions and magnifications could be chosen in real-time, and objects of interest could be highlighted. For presentation, smoothing algorithms were used.

328

Examination analysis Three experienced radiologists (B.B.F., F.C.C., D.H.S.) routinely assessed all axial CT slices and reached a consensus with respect to unexpected pathological findings, such as focal lesions and fatty transformation; the latter is of importance because liver function is decreased in steatosis which was defined as liver attenuation minus spleen attenuation less than or equal to -10 Hounsfield units on non-enhanced CT images [11]. Segmented vessel trees were analysed with respect to vessel architecture, diameters and anatomical variants. Volume of total liver parenchyma, right and left hepatic lobe and left-lateral segments II+III were calculated. All results were discussed with the surgeons of the transplantation team by interactive 3D presentation of all vessels and segmented liver volumes at the workstation in order to evaluate the suitability of potential donors. In order to compare the results of this semiautomatic volume calculation, a conventional liver volumetry first described by Heymsfield et al. [12] and according to the technique described by Kamel et al. [3], was additionally performed. Comparison of intraoperative findings and statistical analyses Depending on vascular anatomy and volumetric analysis, potential donors were excluded from or approved for LDLT. For the 27 patients undergoing transplantation, the anatomical structures were documented in the operation protocol and compared with preoperative findings. The resected liver grafts were weighed immediately after resection, and assuming a specific weight of healthy liver parenchyma of 1 g/ml [13], the association of preoperatively calculated volumes and intraoperatively measured volumes were statistically analysed. For statistical analysis and data management, the SPSS version 10.0 was used. Descriptive statistical measures including mean value, standard deviation and range were calculated for the results of both volumetry methods and the assessment of liver volume after surgery. The association of liver volume and the results of the different volumetry methods was analysed by a correlation analysis (Pearson’s correlation coefficient) and bivariate regression analysis (inclusion method); p values <0.05 were considered significant. Additionally, the results were displayed as scatterplots with adjustment lines. These curves were used to develop the corresponding algorithms rendering the estimation of the liver volume as a function of the semi-automatically calculated liver volumes (calibration curves). Furthermore, the distribution of the differences between semi-automatically calculated liver volume and intraoperatively measured liver volume in relation to their mean values was shown in a scatterplot to check for systematic deviations [14].

Results Vessel segmentation The arterial vascular anatomy was correctly analysed in all 56 potential donors, and in the vast majority of the cases (46 potential donors) the first phase of helical CT allowed an automatic segmentation of hepatic arteries from the respective central vessel (coeliac trunk, superior mesenteric artery or abdominal aorta) to second-order branches and visualisation in a shaded-surface-display mode (Fig. 1). In 10 cases automatic segmentation of the liver arteries was not possible due to small vessel diame-

Fig. 1 Anterior oblique 3D view of segmented vessels within the liver. Arteries (red) are displayed separately. There is a normal arterial supply of the liver via the common hepatic artery arising from the coeliac trunk. No accessory or aberrant arteries are seen

ter (<2 mm) or close proximity to portal veins. In these cases arteries were assessed by multiplanar reformations (MPRs) and maximum intensity projections (MIPs). Among the 56 potential donors, 26 variants of hepatic arteries were found in 22 cases. Most often, accessory or aberrant left hepatic arteries arising from the left gastric artery were found. In a single case an aberrant right hepatic artery arising from the abdominal aorta was detected. The second phase of helical CT always yielded a sufficient contrast between liver parenchyma and hepatic and portal veins. Venous vessels were segmented without difficulties using an automatically proposed threshold and a region-growing algorithm. In 7 cases (12.5%) thresholds were slightly adjusted to achieve a more precise segmentation. For subsequent volume calculation of liver segments, at least fourth-order branches of the portal venous system were segmented; however, 3D visualisations of portal veins were limited to first-, second- and third-order branches to facilitate the visual assessment. Hepatic veins were delineated from the periphery to the inferior vena cava. Thirty-two donors showed 34 variants of hepatic veins, mainly one or more accessory right hepatic veins, and five portal vein trifurcations were found. In the 5 cases where a cholangiographic contrast agent was administered, the main bile ducts were visible on axial slices, but low contrast and small diameters in the periphery prevented a complete 3D visualisation. We refrained from further examinations with the cholangio-

329

ments II+III (n=11). After calculation, the results were presented three-dimensionally (Fig. 2). Data processing time The mean total post-processing time for the first 8 potential donors where an early release of the software was used, was 1 h 47 min (ranging from 1 h 10 min to 2 h 55 min). Using the latest release of the software for the following 48 potential donors, all operators needed a mean total post-processing time of 23 min (range 17–45 min). On average, the various steps required the following processing times: 10 min for the segmentation of liver parenchyma; 4 min for the segmentation of hepatic vessels; 8 min for the analysis of hepatic vessels; and 1 min for the calculation and visualisation of individual portal venous segments. Donor selection Fig. 2 Anterior 3D view of a virtually divided liver and a non-divided portal venous tree. The right lobe (dark magenta) with a calculated volume of 950 ml, accounting for 71% of the total liver volume, was transplanted into a 56-year-old recipient

graphic agent as it increased the liver parenchyma and hindered the segmentation of hepatic arteries and portal veins due to their close contact to the bile ducts. Volumetry Volumes of the whole liver, the right and the left lobe, as well as the left lateral segments II+III, were determined. On average, 92 slices (range 89–97 slices) were reconstructed for each phase of CT imaging. Interactive segmentation of liver parenchyma was performed on every fifth slice of the venous phase. In 9 cases (16%) narrow spacing of the auxiliary points became necessary because of a low contrast between the liver parenchyma and the diaphragm, intercostal muscles or the stomach wall. Eight hundred sixty-two (17%) of the 5159 automatically interpolated contours had to be corrected interactively because of partial inclusion of neighbouring structures. All corrections were minor and took only a few seconds in each case. Volumes of liver segments were automatically calculated based on the individual portal venous branches. The mean volumes of the liver (n=56) were 1680±260 ml (total liver), 1120±210 ml (right liver lobe), 550±130 ml (left liver lobe) and 280±50 ml (left lateral liver lobe, segments II+III). Calculated graft volumes were 1110±180 ml for right lobes (n=15), 820 ml for the left lobe and 270±30 ml for seg-

One potential donor with steatosis hepatis was not approved for LDLT due to an inadequate graft volume. Simple cysts in 13 cases, and small hemangiomas in 4 cases, were detected. These liver lesions had no influence on donor selection. As 2 recipients died prior to transplantation, the number of potential donors decreased from 56 to 54. Twenty-nine potential donors (28 right liver lobes and 1 left liver lobe) were considered for transplantation in adolescent or adult recipients. In 25 cases left-lateral segments were evaluated as grafts for children. Twenty-seven of the 54 potential donors were excluded from transplantation. Since the remaining liver volume of the potential donor after donation was to be at least 30% of the original volume [15], 13 of them were excluded from donation because the remaining liver volume would have been too small. In all of these cases, adult-to-adult transplantations had been planned. For example, a 33-year-old man as a donor to his father would have kept a residual liver volume of 500 ml, which was only 22% of his regular liver volume. Reasons for excluding candidates from adult-to-child donation were a combination of aberrant arteries in 1 patient, and in the remaining 13 cases the virtual graft had an inadequate volume or shape to ensure a sufficient liver function after transplantation. Intraoperative findings and correlation with preoperative evaluation In the 27 transplanted cases, vessel anatomy and variants were documented intraoperatively and compared with the preoperative findings. In 1 case a minor 1-mm-wide

330

Fig. 3A, B Intraoperatively measured graft volumes, conventionally calculated volumes and automatically calculated volumes of A segments II+III and B the right liver lobes. A Intraoperatively measured graft volumes/conventionally calculated volume/automatically calculated volume (n=11). Mean (SD): 290 (30)/270 (30)/280 (50); min–max 240–350/220–300/220–370. B Intraoperatively measured graft volumes/conventionally calculated volume/automatically calculated volume (n=15). Mean (SD): 910 (160)/1110 (190)/1100 (220); min–max 610–1210/850–1530/680– 1500

accessory right hepatic vein originating from segment VI was found that had not been detected on preoperative evaluation. It drained into the inferior vena cava 1.5 cm below the main right hepatic vein. In all other cases the vessel anatomy corresponded with the preoperative findings. Resected liver grafts were weighed intraoperatively in all transplanted cases, and their volumes were calculated based on the specific weight of 1 g/ml. The mean volume of right lobes was 910±160 ml (n=15) and left lateral segments II+III were 290±30 ml (n=11; Fig. 3). The

Fig. 4A, B Close association of the results of the two different volume assessment methods and the intraoperatively measured graft volumes for A segments II+III and B the right liver lobe. There was no significant difference between conventionally and automatically calculated liver volumes. Green markers correlation liver volume, automatically calculated liver volume; red markers correlation liver volume, conventionally calculated liver volume; green lines function of liver volume depending on automatically calculated liver volume; red lines function of liver volume depending on conventionally calculated liver volume

volume of a single left lobe was 630 ml. There was a strong statistical association between intraoperatively assessed graft volume and the preoperatively semi-automatically calculated liver volume by computer-assisted volumetry considering all of the resected grafts. The Pearson correlation analysis showed a correlation of r=0.98 (p<0.001), and the result of the regression model supported this finding (p<0.0001; Fig. 4). The regression analysis investigating the association of graft volume and preoperatively conventionally calculated liver volume revealed a strong correlation as well (p<0.0001; Fig. 4). The scatterplot displaying the distribution of the differences of semi-automatically calculated liver volumes and the graft volumes in relation to the mean value, indicated that with growing liver volume, the differences

331

Fig. 5 Under- and overestimation (dots below and above the zero line) of automatically calculated liver graft volumes. With increasing graft size, a systematic shift towards overestimation and an increasing scattering is noticed. Number of cases=27

were not normally distributed (Fig. 5). The deviations of the estimated volumes in grafts of more than 500 ml showed constant overestimation as a systematic measuring error. Due to this finding, separate algorithms for estimation of liver volume based on semi-automatically calculated liver volumes were generated for grafts £500 ml and those of >500 ml. The equation for liver volume £500 ml reads y=0.31x+190 and for liver volume >500 ml, y=0.54x+200. Postoperative course One of the 27 donors developed a partial thrombosis of the upper mesenteric vein which was confirmed by postoperative US and CT examination. It resolved completely after 2 weeks of heparin infusion therapy followed by 6 weeks of low-dose heparinisation. No further invasive therapy was necessary. All other 26 donors had no postoperative complications. Twenty-four recipients had an uneventful postoperative course. Three recipients developed a haematoma. In one of these patients the haematoma required multiple re-operations, and following a long stay in the intensive care unit, the patient died 78 days after transplantation secondary to multiorgan failure in a septic shock. None of the postoperative complications were associated with inadequate graft size or undetected variants of hepatic vessels.

Discussion Graft size and volumetric analysis In LDLT several requirements concerning liver volume and shape have to be fulfilled in order to avoid postoperative liver failure of the donor and the recipient [15, 16]; therefore, an exact preoperative volume calculation of

both the graft to be transplanted and the remaining liver part of the potential donor is of major importance. Since its introduction in the late 1970s by Heymsfield et al. [12], several new visualisation techniques and refinements in semi-automatic liver volumetry allowing individualised computer-generated resection protocols have been developed [5, 17]. Although the previously reported methods easily allow the volumetry of the total liver parenchyma and the left lateral segments, an arbitrary orientation based on anatomical structures, such as gallbladder, falciforme ligament and hepatic veins, is still necessary for volumetric analysis of other liver segments and the division of the whole liver into its right and left liver lobes. In a recent study Kamel et al. arbitrarily defined the virtual resection plane immediately to the right of the middle hepatic vein, avoiding major vessels traversing between the right and left lobes and extending along the gallbladder fossa anteriorly and the portal bifurcation posteriorly [3]. Although accurate, the procedure was very operator dependent; therefore, new mathematical models are needed. Our study presents a new approach for separating the liver segments by determining the perfusion areas of the individual portal venous branches that constitute the different liver segments. Firstly, we segmented the liver parenchyma using a fast semi-automatic algorithm. The user-steered segmentation was performed on every fifth image slice, and all intermediate contours were interpolated automatically. This method offered a significant reduction of interaction compared with the manual segmentation of all slices [8]. Gao et al. presented a liver segmentation method where the resulting liver borders had to be modified manually in more than 20% of the slices [18]. In our study marginal corrections of the liver borders were necessary in only 17% of the interpolated slices. After segmentation, the total liver surface was generated automatically and partitioned into its segments of individual portal venous supply with the segmental volumes being calculated automatically. In accordance with the literature [19], we found a large variance between the volumes of liver segments. In our study, volumes ranged from 760 to 1770 ml for the right lobe, 330 to 760 ml for the left lobe and 190 to 410 ml for the left lateral segments II+III. In the present investigation, the comparison of preoperatively semi-automatically calculated liver volume by computer-assisted volumetry and the intraoperatively measured graft volume revealed a tight correlation; however, a systematic deviation of calculated liver volume and measured liver volume was observed in grafts of >500 ml in terms of constant overestimation. Other studies reported in the current literature show acceptable correlations between estimated and measured liver volume as well [4, 13, 19, 20]. Closer analysis of these studies showed differing deviations, such as systematic underestimation of left-lobe volume [19] and a high variance of the measurements [13, 20].

332

The systematic overestimation of larger liver grafts reported in the present study can be explained primarily by a higher blood content in these grafts in the larger liver vessels and the liver parenchyma on the sinusoidal level as well. Since the resected grafts had lost blood to a relevant extent prior to weight assessment, a difference in comparison with the volumetry of the physiologically perfused liver was to be expected. This would affect especially the right and left liver lobes because of their large blood volumes. In a recent experimental study (pig model) the descibed effects on volumetry measurement have been demonstrated (B.B.J. Frericks et al., unpublished data). Another possible reason for differences of estimated and measured graft volume could be deviations of virtual and operative resection plane. But since visualisation problems would particularly affect the assessment of smaller structures and deviations would occur as under- and overestimations, it is not likely that this methodological issue played a major role in the systematic overestimation of the larger liver volumes in the present analysis. Deviations of volumetry estimations can be found in all of the currently employed assessment methods and there is no accepted preoperative gold standard for a comparison of measurements. In the present investigation an approach was developed to correct for the detected systematic estimation error. According to these equations, representing calibration curves of the correlation graphs, the real graft volume can be calculated based on the volumetric estimation. Presentation and donor selection After segmentation and volumetric analysis, the segmented liver and the hepatic vessels were visualised three-dimensionally. Various viewing angles were chosen interactively in real-time. Surface and transparency values were assigned for the design of individual presentations. Annotations, which informed the user about the volume of specific structures, e.g. were easily displayed. All cases were presented and discussed in an interdisciplinary transplantation conference. Based on the results of the preoperative evaluation, 50% of the potential donors were excluded from LDLT. In one case, the potential donor was excluded due to variants of vessels. In 26 other cases, the potential donor was excluded due to an inadequate volume of the remaining liver or the liver graft, or an inadequate shape of the liver graft. The exclusion rate of 50% corresponds to the exclusion rates of other transplantation centres, although lower, but even higher rates have been reported. We assume that our careful preoperative evaluation prevented futile laparotomies, whereas the possibility of falsely excluding potential donors is very low.

Limitations A limitation of all investigations evaluating the accuracy of hepatic volumetry in a clinical setting, especially for segment oriented liver parts, as in our study, is that there is no gold standard; however, the requirements for clinical volumetry, i.e. reproducibility, reliability and processing speed, are adequately met with this new mathematical approach. Since the patient selection was based on the results of the liver segmentation software, there might be statistical bias due to the diagnostic review bias or the verification bias. The final outcome parameter (liver volume) might be influenced by the interpretation of the prior test (donor selection by volumetry), or the prior test result might influence the application of the gold standard. But since there is no possibility of assessing liver volume in nonresected patients, and there is no alternative treatment option except for non-treatment, it is not possible to control for this statistical problem. This methodological issue has been a limitation of the statistical evaluations of all other current investigations on liver volumetry as well. The mean time span between pre-operative CT scan and transplantation of 41 days seems to be quite long; however, this is explained by the highly elective circumstances in LDLT in most cases. There were 11 cases where the transplantation was performed within 2 weeks and 2 cases where the transplantation was performed on the day following the CT scan. Sufficient automatic segmentation and 3D visualisation of the hepatic arteries was possible in only 46 of 56 potential donors. This could have been improved using a 1.25-mm collimation or less. With the new 8- and 16row CT scanners a complete thin-slice-coverage liver scan is possible, and this will further improve the hepatic artery visualisation; however, except for one non-visualised small accessory hepatic vein, the predicted vessel anatomy corresponded to the intraoperative findings in all 27 cases, and based on MPRs and MIPs, the hepatic artery anatomy was correctly predicted in all 27 transplanted cases. Although comparable to other reports concerning CTbased liver volumetry [21], the post-processing time of 23 min appears to be quite long; however, it is justifiable, because a semi-automatic segmentation of the liver parenchyma is performed on all of the mean 92 thin slices (slice thickness 2 mm) of the liver. Based on the segmented hepatic vessel architecture the different liver segments are calculated and visualised automatically and operator independently. Taken together, these elaborated mathematical models may result in longer post-processing times, on the one hand, but allow a more reproducible liver segmentation, on the other. This may not only be of interest in LDLT but also in other clinical settings in general hepatic surgery; however, the extended time

333

and personnel requirements have to be taken into account when considering the procedure for introduction into the clinical routine.

Conclusion In conclusion, this study shows a new technique of semiautomatic segmentation of the human liver based on a mathematical model which considers the portal venous blood supply of the different liver segments rather than

peripheral landmarks. It allows accurate semi-automatic volume calculation of various segments. The results of this preoperative evaluation protocol influenced the donor selection process and enabled a three-dimensional display of the vessels and the parenchyma with delineation of the segments, which simplified the three-dimensional orientation for the transplantation surgeons. We believe that the technique of perfusion-based liver segment volumetry presented in this study could also be very helpful in the pre-operative planning and risk assessment in general liver surgery.

References 1. Emond JC, Renz JF, Ferrell LD, Rosenthal P, Lim RC, Roberts JP, Lake JR, Ascher NL (1996) Functional analysis of grafts from living donors: implications for the treatment of older recipients. Ann Surg 224:544–554 2. Ishifuro M, Horiguchi J, Nakashige A, Tamura A, Marukawa K, Fukuda H, Ono C, Akiyama Y, Kushima T, Ito K (2002) Use of multidetector-row CT with volume renderings in right lobe living liver transplantation. Eur Radiol 12:2477–2483 3. Kamel IR, Kruskal JB, Warmbrand G, Goldberg SN, Pomfret EA, Raptopoulos V (2001) Accuracy of volumetric measurements after virtual right hepatectomy in potential donors undergoing living adult liver transplantation. AJR 176:483–487 4. Schroeder T, Nadalin S, Stattaus J, Debatin JF, Malagó M, Ruehm SG (2002) Potential living liver donors: evaluation with an all-in-one protocol with multidetector-row CT. Radiology 224:586–591 5. Lamade W, Glombitza G, Fischer L, Chiu P, Cardenas CE, Thorn M, Meinzer HP, Grenacher L, Bauer H, Lehnert T, Herfarth C (2000) The impact of 3-dimensional reconstructions on operation planning in liver surgery. Arch Surg 135:1256–1261 6. Selle D, Spindler W, Schenk A (2000) Computerized models minimize surgical risk. Diagn Imaging Europe 12:16–20 7. Schenk A, Prause G, Peitgen HO (2001) Local cost computation for efficient segmentation of 3D objects with Live Wire. Medical imaging: image processing. Proc SPIE 4322:1357–1364

8. Schenk A, Prause G, Peitgen HO (2000) Efficient semiautomatic segmentation of 3D objects in medical images. Medical image computing and computer-assisted intervention (MICCAI 2000). Springer, Berlin Heidelberg New York, pp 186–195 9. Selle D, Peitgen HO (2001) Analysis of the morphology and structure of vessel systems using skeletonization. SPIE Med Imaging 4321:271–281 10. Selle D, Preim B, Schenk A, Peitgen HO (2002) Analysis of vasculature for liver surgical planning. IEEE Trans Med Imaging 21:1–15 11. Alpern MB, Lawson TL, Foley WD, Perlman SJ, Reif LJ, Arevalos E, Rimm AA (1986) Focal hepatic masses and fatty infiltration detected by enhanced dynamic CT. Radiology 158:45–49 12. Heymsfield SB, Fulenwider T, Nordlinger B, Barlow R, Sones P, Kutner M (1979) Measurement of liver, kidney and spleen volume and mass by computed axial tomography. Ann Intern Med 90:185–187 13. Van Thiel DH, Hagler NG, Schade RR, Skolnick ML, Heyl AP, Rosenblum E, Gavaler JS, Penkrot RJ (1985) In vivo hepatic volume determination using sonography and computed tomography: validation and comparison of the two techniques. Gastroenterology 88:1812–1817 14. Bland JM, Altman DG (1986) Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 8476:307–310 15. Fan ST, Lo CM, Liu CL, Yong BH, Chan JK, Ng IO (2000) Safety of donors in live-donor liver transplantation using right lobe grafts. Arch Surg 135:336–340

16. Kiuchi T, Kasahara M, Uryuhara K, Inomata Y, Uemoto S, Asonuma K, Egawa H, Fujita S, Hayashi M, Tanaka K (1999) Impact of graft-size mismatching on graft prognosis in liver transplantation from living donors. Transplantation 67:321–327 17. Luccichenti G, Cademartiri F, Cobelli R, Pavone P (2003) Assessment of organ volume with different techniques using a living liver model. Eur Radiol 13:1286–1290 18. Gao L, Heath DG, Kuszyk BS, Fishman EK (1996) Automatic liver segmentation technique for threedimensional visualization of CT data. Radiology 201:359–364 19. Kawasaki S, Makuuchi M, Matsunami H, Hashikura Y, Ikegami T, Chisuwa H, Ikeno T, Noike T, Takayama T, Kawarazaki H (1993) Preoperative measurement of segmental liver volume of donors for living related liver transplantation. Hepatology 18:1115– 1120 20. Schiano TD, Bodian C, Schwartz ME, Glajchen N, Min AD (2000) Accuracy and significance of computed tomographic scan assessment of hepatic volume in patients undergoing liver transplantation. Transplantation 69:545–550 21. Sandrasegaran K, Kwo PW, DiGirolamo D, Stockberger SM, Cummings OW, Kopecky KK (1999) Measurement of liver volume using spiral CT and the curved line and cubic spline algoriths: reproducibility and interobserver variation. Abdom Imaging 24:61–65