Fast and accurate liver volumetry prior to hepatectomy

Abstract

Background

Volumetric assessment of the liver is essential in the prevention of postresectional liver failure after partial hepatectomy. Currently used methods are accurate but time-consuming. This study aimed to test a new automated method for preoperative volumetric liver assessment.

Methods

Patients who underwent a contrast enhanced portovenous phase CT-scan prior to hepatectomy in 2012 were included. Total liver volume (TLV) and future remnant liver volume (FRLV) were measured using TeraRecon Aquarius iNtuition^® (autosegmentation) and OsiriX^® (manual segmentation) software by two observers for each software package. Remnant liver volume percentage (RLV%) was calculated. Time needed to determine TLV and FRLV was measured. Inter-observer variability was assessed using Bland-Altman plots.

Results

Twenty-seven patients were included. There were no significant differences in measured volumes between OsiriX^® and iNtuition^®. Moreover, there were significant correlations between the OsiriX^® observers, the iNtuition^® observers and between OsiriX^® and iNtuition^® post-processing systems (all R² > 0.97). The median time needed for complete liver volumetric analysis was 18.4 ± 4.9 min with OsiriX^® and 5.8 ± 1.7 min using iNtuition^® (p < 0.001).

Conclusion

Both OsiriX^® and iNtuition^® liver volumetry are accurate and easily applicable. However, volumetric assessment of the liver with iNtuition^® auto-segmentation is three times faster compared to manual OsiriX^® volumetry.

Introduction

Postresectional liver failure (PLF) remains the most important factor associated with postoperative mortality after major liver resections (resection of 4 or more Couinaud liver segments).^{1, 2, 3} Prevention of this severe and often lethal complication is attempted through pre-operative CT-based volumetric liver assessment in patients undergoing major liver resections.^{4, 5, 6} In patients with healthy livers approximately 25% of the liver parenchyma needs to be preserved to prevent PLF.⁷ In damaged or cirrhotic livers up to 50% liver parenchyma needs to be spared.

The standard technique for liver volumetry is CT based (semi-) manual delineation of liver borders. This volumetric assessment is relatively time-consuming.^{4, 5, 6} Additionally, with traditional software, the total liver volume, future remnant liver volume and tumour volume all need to be outlined separately in order to make a proper risk-assessment prior to large liver resections, further increasing the time required for complete analysis.^{4, 6}

With specialised software, auto-segmentation of organs is possible.^{8, 9, 10, 11, 12, 13, 14} However, CT-based automated segmentation of the liver is challenging because organs with comparable density surround the liver. Automated future remnant liver volumetric analysis has not yet been validated by comparing it with manual slice-by-slice drawing of contours. Recently, automated liver segmentation has been introduced in a new version of TeraRecon Aquarius iNtuition^® which is user friendly and easy to learn. Moreover, a ‘virtual cut’ option is available, which is timesaving as it directly provides the future remnant liver volume after performing the cut on the reconstructed total liver volume.

This study aimed to validate auto-segmentated liver volumetry using the iNtuition^® software and investigate the time saved with this new method compared to semi-manual outlining of the liver contours using traditional software (OsiriX^®).

Methods

Patients

Thirty patients who underwent liver surgery in 2012, in whom a contrast enhanced portovenous phase CT-scan of the liver had been performed as part of their preoperative assessment, were retrospectively included in the present study. The maximum CT-scan slice thickness was 3 mm. For standardisation reasons, all volumes were measured as if patients would undergo a right extended hemihepatectomy including resection of the caudate lobe (Couinaud segments 4–8 + 1), regardless of the location of the tumour and the surgical plan. Moreover, tumours were not outlined.

Volumetric liver analysis using OsiriX^®

Liver volumetry based on manual delineation of the liver with for instance OsiriX^® is an accurate method to measure total liver volume (TLV) and predict future remnant liver volume (FRLV)⁴ and is considered the gold standard.^{15, 16} Therefore, OsiriX^® liver volumetry was selected as reference measurement to compare with auto-segmentation using iNtuition^®. Volumetric analysis using OsiriX^® version 5.0.2 32-bit was performed by two observers (observer 1 and 2). Observer 1 was experienced with OsiriX^® while observer 2 was new to post-processing software (TML and KMCvM; observers 1 and 2, respectively). Both observers were PhD candidates focussing on liver surgery. Prior to the evaluation of the studies, observer 1 instructed observer 2 how to use the OsiriX^® software on scans which were not included in the present study. Details on how volumetry was performed with OsiriX^® have been described previously.⁴ TLV and FRLV, i.e. Couinaud segments 2 and 3, were semi-manually outlined on a pre-operative, contrast enhanced, portovenous phase CT scan. As the falciform ligament separates the virtual liver remnant from segment 4 and 1, this landmark was used to indicate the virtual transection on the CT scan. A marker was placed between the falciform ligament at the base of segment 3 (caudally), and the fork of the middle and left hepatic vein (cranially). When segments 2 and 3 were not visually separated from segment 4 and 1 by the falciform ligament, the marker was used to point out the virtual transection (Fig. 1a).

Figure 1.

a. OsiriX^® semi-manual volumetric assessment, with reconstruction (right bottom); b. INtuition^® auto-segmentation with “virtual cut” function (left bottom), reconstruction (right bottom)

Volumetric liver analysis iNtuition^®

Liver volumetry using iNtuition^® was also performed by two observers (observer 3 and 4). Observer 3 had experience with iNtuition^® and other post-processing software, while observer 4 only had limited experience with post-processing software in general and none with iNtuition^® (CWKPA and MJL, respectively). Both observers were radiologists. To ensure that it would not be the radiologic background and experience of the iNtuition^® observers but the software itself would be responsible for possible timesaving, observer 1 (TML) later on also performed all volumetries with the iNtuition^® software. Prior to evaluation of the studies, observer 3 instructed observer 1 and 4 how to use the iNtuition^® software on scans which were not included in this study.

After loading the CT scan into the iNtuition^® software, the LD2 module was selected from the tools menu. Once open, the ‘segment’ tool was selected and auto-segmentation performed. After calculation was completed, manual evaluation of the outlined contours was performed and adjustments were made with the ROI selection tools if needed. This volume was then captured using the ‘capture’ tool and rendered. After continuing the process, the TLV was automatically calculated. Next, using the ‘virtual-cut’ tool, the virtual extended right hemihepatectomy with the previously described landmarks was performed, using axial and coronal slices as references. The RLV was then calculated automatically by selecting the ‘cut’ tool (Fig. 1b). All observers were blinded for volumetric results of the other 3 observers.

Timing of volumetry

Time needed for assessment of the TLV, the FRLV, and complete volumetric liver assessment were measured with a stopwatch. Time to calculate the TLV was measured from the moment of the first contour drawing with OsiriX^® or start of auto-segmentation with iNtuition^® until calculation of the TLV. Time to calculate the FRLV was measured from the moment of the first contour drawing with OsiriX^® or start of virtual cut with iNtuition^® until calculation of the remnant liver volume. Time needed for complete volumetric liver assessment was calculated by adding the time necessary for TLV calculation to the time needed for FRLV calculation.

Statistics

TLV, FRLV, calculated FRLV as percentage from TLV (RLV%), and duration of measurements are presented as mean ± standard deviation (SD). Pearson's correlation coefficient R² was used to test correlation between measurements. The mean of volumes measured by the observers using OsiriX^® software was assumed to be the reference standard⁴ to validate iNtuition^® volumetry. The differences between observers 1 and 2 (OsiriX^®), 3 and 4 (iNtuition^®), and mean of observers 1/2 versus the mean of observers 3/4 (OsiriX^® versus iNtuition^®), were tested using one-way ANOVA and paired t-tests using the Bonferroni correction. Interobserver variability was assessed using Bland-Altman plots and 95% limits of agreement were provided.¹⁷ A p value < 0.05 was considered to indicate statistical significance. Statistical analysis was performed using the Statistical Package for the Social Sciences (SPSS) version 20.0 and Prism 5.0 (Graphpad software, Inc, San Diego, CA).

Results

Patients

The pre-operative CT-scans of 30 patients were evaluated. Three scans had to be excluded due to inability to perform auto-segmentation. Therefore, 27 patients were analysed in the present study. Details of patient selection and reasons for exclusion are listed in Fig. 2. Median age was 66 [42–81] years and the majority of patients were male (n = 18, 69%). Most of the included patients were diagnosed with colorectal cancer liver metastases (n = 18, 69%). Other indications were cholangiocarcinoma (n = 2, 8%), hepatocellular carcinoma (n = 1, 4%), metastases from carcinoid tumour (n = 1, 4%), and benign lesions (n = 4, 15%).

Figure 2.

Flowchart of the study

Liver volumetry

Interobserver variability using OsiriX^®

TLVs, estimated FRLVs, and RLV% of all observers are depicted in Fig. 3. Observers 1 and 2 showed excellent volumetric agreement for TLV (R² = 0.998, p < 0.001), FRLV's (R² = 0.997, p < 0.001) and RLV% (R² = 0.993, p < 0.001) (Table 1). Moreover, excellent interobserver variability was found between TLV, FRLV, and RLV% measurements of observer 1 and 2 (Fig. 4).

Figure 3.

Mean total liver volumes (TLV), future remnant liver volumes (FRLV) and remnant liver percentages plotted for all observers. Horizontal lines represent mean ± SD

Table 1.

Volumetric assessment and duration

	Observer 1. OsiriX	Observer 2. OsiriX	Observer 3. INtuition	Observer 4. INtuition
Volumes
TLV, mLs	1717 ± 362	1741 ± 369	1778 ± 388	1754 ± 382
FRLV, mLs	344 ± 157	345 ± 155	355 ± 151	334 ± 146
RLV, %	20.0 ± 7.1	19.8 ± 7.1	20.0 ± 7.0	19.0 ± 6.5
Time needed
Measurement TLV, min	10.7 ± 2.8	15.3 ± 5.3	4.0 ± 1.3	5.7 ± 2.7
Measurement FRLV, min	4.5 ± 1.2	6.2 ± 2.4	1.0 ± 0.5	0.9 ± 0.6
Total assessment, min	15.2 ± 3.5	21.5 ± 7.4	5.0 ± 1.6	6.6 ± 2.7

Figure 4.

Interobserver variability using OsiriX^® (left) and iNtuition^® (right). Left: mean total liver volumes (TLV), future remnant liver volumes (FRLV), and remnant liver volume % of observers 1 and 2 are plotted against the difference between observers 1 and 2. Right: mean TLV, FRLV, and RLV% of observers 3 and 4 are plotted against the difference between observers 3 and 4

Interobserver variability using iNtuition^®

Observers 3 and 4 also showed excellent volumetric agreement, R² = 0.995, p < 0.001, R² = 0.979, p < 0.001 and R² = 0.993, p < 0.001 for TLV, FRLV and RLV% respectively. Excellent interobserver variability was found between the 2 iNtuition observers^® (Fig. 4).

Interobserver variability between OsiriX^® and iNtuition^®

No differences were found when comparing the means of the liver volumes as assessed by the OsiriX^® observers with those assessed by the iNtuition^® observers. TLV's were 1729 ± 365 and 1760 ± 377 (R² = 0.997, p < 0.001), FRLV's were 345 ± 156 and 345 ± 147 (R² = 0.985, p < 0.001), and RLV% were 19.9 ± 7.1 and 19.5 ± 6.7 (R² = 0.984, p < 0.001) respectively measured with OsiriX^® and iNtuition^®. Good interobserver variability between OsiriX^® and iNtuition^® liver volumetry was found (Fig. 5).

Figure 5.

Interobserver variability using OsiriX^® vs iNtuition^® (observers 1/2 vs 3/4). OsiriX^® and iNtuition^® volumetry of TLV, FRLV and RLV% is plotted against the difference between OsiriX^® and iNtuition^® (left). Mean TLV, FRLV and RLV% measured with OsiriX^® plotted against the same volumes measured with iNtuition^®

Liver volumetry by a non-radiologist (observer 1) using iNtuition^® was comparable to the means of the Osirix^® measurements with a mean TLV of 1778 ± 372 (R² = 0.986, p < 0.001), a mean RLV of 363 ± 155 (R² = 0.967, p < 0.001) and a RLV% of 20.3 ± 6.9 (R² = 0.961, p < 0.001).

Duration of assessment

Time needed for assessment of TLVs, estimated FRLVs, and RLV% of all observers are depicted in Fig. 6. Median time needed for total volumetric assessment (TLV + FRLV) with OsiriX^® and iNtuition^® was 18.4 ± 4.9 and 5.8 ± 1.7 min respectively (p < 0.001). Time needed for measurement of the TLV was over 2.5 fold reduced (13.0 ± 3.6 vs 4.9 ± 1.8 min, p < 0.001) when iNtuition^® was used compared with OsiriX^®. Using iNtuition^® the FRLV assessment was even reduced over 5 times (5.3 ± 1.6 vs 1.0 ± 0.4 min, p < 0.001) (Table 1).

Figure 6.

Time needed for total volumetric assessment of the liver, measurement of TLV, and FRLV for all observers

The time needed by the non-radiologist using the iNtuition^® software was comparable to the time needed by the radiologists using iNtuition^®. Time needed for TLV, RLV and total assessment was 5.2 ± 3.1 (p = 0.444), 1.0 ± 0.3 (p = 0.665) and 6.2 ± 3.2 (p = 0.370), respectively.

Discussion

In this study, two methods of liver volumetry were compared to analyse whether there is an advantage in using auto-segmentation of the liver instead of manual outlining. Manual outlining of liver borders for preoperative volumetric evaluation of the liver is time-consuming and has been performed in recent years with specific post-processing software such as OsiriX^® or Image J^®.^{4, 5, 6} The present study showed that volumetry of the liver using Terarecon Aquarius iNtuition^® auto-segmentation was three times faster than manual outlining of the liver with OsiriX^®, while maintaining accuracy.

Two observers in the present study were experienced and well trained in performing liver volumetry. The other 2 observers (one for each post-processing system) were newly-trained by one of the 2 experienced observers. The present study showed that after a short training liver volumetry could be performed adequately, even in inexperienced hands. Others already showed that CT based auto-segmentation of the TLV was also possible with Windows 4.4 framework^®,⁸ Philips Intellispace Portal^®,⁹ Voxar^®,¹² and CAD technologies^®.¹³ Only two studies actually compared manual with automated volumetry with generally good correlations.^{8, 13} These studies were done on liver transplant patients, and only total liver volumetry was performed. To our knowledge the present study is the first to not only test total liver volumetry but also future remnant liver volumetry and determination of the remnant liver volume percentage. In addition, the time needed for auto-segmentation and volumetry of the total liver volume with iNtuition^® seems comparable to other post-processing software packages, showing times ranging from <1 min to 5 min compared to less than 5 min in the present study.^{8, 9, 13}

The potential advantage of CT based auto-segmentation with iNtuition^® over other post-processing systems consists of the ‘virtual cut’ option which enables fast and accurate determination of the future remnant liver volume. Not only can this be done on the segmented volume and the multiplanar reconstructions in a straight line, it can also be adjusted in all directions following the exact contours of the planned resection. In the previous decade the number of patients found eligible for partial liver resection and the extent of liver resections increased.^{18, 19} By performing more major liver resections, the risk of leaving a too small liver remnant behind became more prominent and liver volumetry gained more attention.^{18, 19} In patients undergoing large liver resections it is essential that the surgical plan is precisely outlined in order to make a proper pre-operative risk assessment and prevent PLF. Most frequently radiologists perform liver volumetry. However, a transection in patients needing large and possibly critical liver resections is often not straightforward and cannot always be translated into an anatomical liver resection as described in the IHPBA Brisbane nomenclature.²⁰ In these cases the surgical eye and expertise are essential to precisely indicate the future remnant liver in order to evaluate the risk of PLF. At a time in which patients are preoperatively discussed during multidisciplinary meetings, fast and reliable volumetry during such meetings seems logical for patients planned for major liver resections. Theoretically, radiologists could perform TLV volumetry in patients with possible critical resections prior to these multidisciplinary meetings which, as we have shown in the present study, is only an additional burden of about 5 min to their work. Together with the surgeon's input the radiologist could then perform the virtual cut during the multidisciplinary meeting within 1 min, providing useful information concerning the safety of the particular surgical plan. Also, as PLF still occurs,^{1, 2, 3} we believe that standard future remnant liver percentage calculations should be incorporated into all radiological reports of patients planned for major liver resections.

The advantage of post-processing software that communicates with a PAC system is that the surgeon might adjust the virtual cut himself behind his desk. The surgeon could potentially even adjust the virtual cut in the operating theatre when the surgical plan needs to be adjusted intra-operatively due to more widespread disease than anticipated preoperatively. The future role of post-processing open source software such as Osirix^® is unclear as the time needed for volumetric assessment of the liver by non-radiologists was almost 20 min in our and sometimes more in previous studies.^{4, 5, 6, 11, 13} To our knowledge, there is no ‘liver plugin’ available in OsiriX^® or other open source post-processing software that allows for automated liver segmentation.

The most important prerequisites for successful auto-segmentation using iNtuition^® are thin slices (1–3 mm) and adequate homogeneous contrast enhancement of the liver parenchyma. In our experience, only minimal manual adjustments are required after the auto-segmentation to exclude some adjacent structures such as ascending colon, ribcage, or a part of the right atrium, if the CT-scans meet the proposed standards. However, the auto-segmentation does not always work well. For instance in one case, the liver tumours bordered directly to the ribcage resulting in inclusion of the right lower ribcage into the volume. This required manual adjustment. On the other hand, once the TLV is calculated, the option to perform a virtual cut is a great advantage.

The FRLV in all patients was well defined in the present study, providing reliable comparison between the post-processing systems. However, due to the standardized virtual resections, we were unable to compare FRLV measured with OsiriX^® or iNtuition^® volumetry with the surgical resection specimens. The CT-scan had to meet specific standards, such as a maximum slice thickness of 3 mm and sufficient contrast enhancement, resulting in a set of CT-scans that were comparable in quality. This can also be seen as a downside, since the quality of CT-scans that are offered from referral centers often do not meet these standards. With our reading design and by having observer 1 perform analysis with both OsiriX^® and INtuition^® the present study shows that the software itself and not the radiologic expertise (radiologists versus liver surgery PhD candidates) saved time.

In conclusion, Terarecon Aquarius iNtuition^® liver volumetry is valid, accurate, and easily applicable. Moreover, auto-segmentation with iNtuition^® is three times faster than (semi-) manual segmentation with OsiriX^® volumetry. Since, in daily practice, time for liver volumetry is limited, a fast (and thus automated) volumetry technique is required to incorporate the findings in radiological reports and multidisciplinary meetings. This would be a valuable addition to image evaluation and may increase the safety of major liver surgery.

Conflicts of interest

None to declare.