Scientific Papers

Can 3D imaging modeling recognize functional tissue and predict liver failure? A retrospective study based on 3D modelling of the major hepatectomies after hepatic modulation | BMC Surgery

This is a retrospective study carried out at Sanchinarro University Hospital, Madrid from May 2015 to October 2019, analysing patients with insufficient FLR volume that required modulation before major hepatectomy. Associated liver partition and portal vein ligation (ALPPS) and portal vein embolisation (PVE) followed by major hepatectomy were the technique used.

For each patient, a 3D imaging reconstruction (3D-MSP®) was performed before and after the liver regeneration technique and the data of anatomical and function FLR volumetry were analysed.

Sanchinarro University Hospital is a reference centre in hepatobiliary surgery with extensive experience in major hepatectomies and especially in the ALPPS technique, which has been described in previous reports without open and robotic approach [4, 5].

Patients selection

Inclusion criteria were patients with unilateral or bilateral liver malignancies with a FLR to total liver volume (TLV) ratio (FLR/TLV) < 25% and FLR/TLV < 40% when liver parenchyma damage (steatosis, fibrosis or chemotherapy-induced liver damage) was suspected in the radiological and laboratory tests. We included patients with Child A liver function and patent right portal vein.

Functional assessement of patients candidate to surgery consisted on assessment of the degree of ascites and evaluation of the presence and severity of encephalopathy. Bioquemical assessment including the levels of bilirubin and albumin and the prothrombin time. Patients were stratified according to the Child Pugh classification.

Exclusion criteria were signs of portal hypertension, such as ascites, and/or intra-abdominal varices, the presence of distant metastasis, and complete right portal vein thrombosis.

The preoperative study consisted of tumour markers (CEA, CA19-9), computed tomography (CT) scans, magnetic resonance imaging (MRI) and positron emission computed tomography (PET TC) and/or positron emission magnetic resonance imaging (PET MRI).

PHLF was defined as the postoperative deterioration of liver function with an increase in the INR and concomitant hyperbilirubinemia on or after postoperative day 5, as proposed by the International Study Group of Liver Surgery (ISGLS) [6]. Postoperative complications were recorded and categorized according to the modified Clavien-Dindo classification [7].

Acquisition and processing of 3D Cella Medical Solutions (3D-MSP®) imaging

A 3D Cella Medical Solutions (3D-MSP®) imaging reconstruction was created for each patient after liver regeneration techniques. In patients undergoing ALPPS, this was performed preoperatively, and one week before the second surgical procedure. In patients undergoing hepatic embolisation, this was performed preoperatively, and one week before the scheduled hepatectomy.

3D Imaging acquisition methods

Data acquisition

all available preoperative images of the patient (CT, MRI, PET-CT, PET-MRI) are used to analyze tumour distribution, estimate remaining liver volume, and identify tumour-vessel relationships with vascular anatomy. An image acquisition protocol is used to normalize the characteristics of the acquired images (3dMSP) (Video 1). The data are obtained in DICOM format (Digital Imaging and Communication in Medicine) and are anonymized.

In this study, CT images only were used in order to prevent the incorporation and fusion of other methods from altering the volumetric results.

Image fusion

The different modelled elements are demarcated in the most appropriate diagnostic imaging sequences. Therefore, the use of image fusion techniques is necessary to correct errors resulting from breathing, movement, position, etc., in the patient. Rigid registration techniques are used for image rotation, translation and scaling alignment, while similar, non-rigid registration techniques are used for 3D imaging for tissue deformation correction.

Image pre-processing and segmentation

Hepatic parenchyma, CVI, supra-hepatic vein, portal vein, hepatic artery, bile duct, and tumour are segmented. If necessary, other structures, such as cysts, hilar adenopathies, prostheses, staples, drainages, etc., are also reconstructed. Noise is previously reduced with anisotropic diffusion filters and N3 algorithms.Convolutional networks and advanced medical image processing techniques as active contour and region growth are used for segmentation.

Processing of modelling

Laplacian smoothing filters are used in the 3D reconstruction of the model to correct the scaling derived from the thickness of the scan. In addition, techniques are used to divide the parenchyma into hepatic segments I-VIII and to subdivide the vascular elements: supra-hepatic in LHV, MHV and RHV; port in RPV, LPV and MPV; artery in LHA, RHA, CHA-Aorta; bile ducts in the gallbladder, cystic, common bile duct, common liver and bile ducts. The models are processed to allow the execution of regulated and unregulated resections, virtual ablations and safety margins of 5, 10 and 15 mm, obtaining the remaining and tumour volumes. Finally, algorithms were developed for the identification of portal and arterial anatomical variants based on atlases and other 3D pattern recognition.

Liver function analysis

Once the imaging studies have been analysed according to inclusion criteria, the segmentation of the structures of interest for analysis is conducted. In this case, the liver parenchyma, arterial and venous vasculature, portal vasculature, tumour regions, and gallbladder are segmented. With the aim of avoiding the impact of vascular and tumour tissues on the acquired textures, radiomic feature extraction of the radiomic characteristics of the liver parenchyma is performed exclusively, removing the vascular and tumour regions from the analysis. These annotations of the structures of interest was performed in both pre-diagnosis CT images and pre-surgery CT images.

Thus, considering ALPSS and embolisation patients as a whole group, radiomics features were extracted on FLR tissue and analysed blindly in all patients. In this way, the goal was to reach the characteristics that could be associated algorithmically to find similarities among all pre-diagnosis studies (considering the parenchyma as functional tissue) and differences with hypertrophied tissue. This provided the features that seemed to be indicators of the differentiation between functional and oedema tissue after hypertrophy on pre-surgery images.

The workflow of the algorithm is showed in Fig. 1.

Fig. 1
figure 1

Workflow of radiomics algorithm

The software to extract and analyse the features was developed in-house based on PyRadiomics library, validated by the IBSI imaging biomarker standardisation initiative (current reference standard for the development of radiomic studies. A total of 106 first order (histogram/intensity-based), shape, and second and higher order (texture-based) features are extracted for each of the imaging studies selected in the analysis. A standardisation of the extracted characteristics is conducted.

Due to the high number of extracted characteristics, it is necessary to conduct a selection of those that contain relevant information for the analysis. In this preliminary study, the information gain metric is used to select the characteristics of first order that add greater value to the analysis. The values of the selected characteristics are analysed through box plots. Additionally, the Student t-test statistical test (assuming normal distributions) was used to compare the mean values of each radiomic characteristic for both study groups. Additionally, an unsupervised clustering method (k-Means) is evaluated for the identification of possible subgroups of studies based on the activation profiles of the radiomic characteristics. Finally, the performance of a multivariate logistic regression model based on selected radiomic characteristics for the classification of liver functional volume above or below the defined threshold based on the texture profile of the liver parenchyma is evaluated.

General requirements for the acquisition of CT images

Acquisition window

All anatomical structures of interest should be visualized

Scan resolution

Distance between scans ≤ thickness (contiguous scans). Distance between scans: max 30% overlap

Image size / Pixel size

– Constant image and pixel size in each sequence

– Minimum size of the matrix 512 × 512

Gantry inclination

Without inclination

Breathing phase

It must be the same in all explorations

Other Recommendations

– Avoid motion artifacts and achieve light, relaxed breathing

– Avoid unnecessary metal artifacts

– Acceptable image disturbance

– Volume of intravenous contrast medium that provides a correct visualization of the structure of interest

Archive format


General requirements for MR imaging

Acquisition window

All anatomical structures of interest should be visualized

Cut resolution

– Distance between cuts ≤ cut thickness (cuts


– Distance between cuts ≤ 6 mm

Image Size / Pixel Size

– Constant image and pixel size within each sequence

– Minimum size of the Matrix 512 × 512

Other recommendations

– Gentle breaths without sudden movements

– Uniform signal on all anatomical structures

– Tolerable image noise

– For tumor localization, an examination protocol is required that produces a sufficiently high contrast for this. It is advisable that you always include a fat suppression sequence (STIR, SPAIR, or similar)

– For images of vascular anatomy, the acquisition of dynamic images with contrast in 2 or 3 times is recommended

Archive format


Specific requirements for MR imaging in 3D-MSP Hepatic Surgery

Dynamic sequences

– Cutting distance: ≤ 6 mm

– Acquisition of dynamic images in 2 times arterial and venous) or in 3 times (arterial, venous, late venous)

– Use bolus tracking

Fat suppression sequences

– Cutting distance: ≤ 6 mm

– For the localization of tumors

– STIR, SPAIR or similar

MRI cholangiography

– Cutting distance: ≤ 6 mm

– Sections oriented axially or coronally (axially recommended)

3D liver volume analysis

The following volumetric were calculated in the 3D model:

  • TV, the tumour volume

  • The total liver volume (TLV) was calculated before starting the regeneration procedure

  • The Anatomic Future Remnant Liver (FRL: hypertrophy + oedema) was calculated by manual delineation according to Couinaud’s functional segmentation of the liver [8]. FRL is the anatomic volumetric value, it includes true hypertrophy tissue + edema/congestion

  • The Functional Future Liver Remanent (FRFx: real hypertrophy) is the real hypertrophy, functional tissue and it was calculated with version 1.2.1 and was used for the first-order liver structure analysis of the algorithm for quantifying and oedema-congestion.

The volume of the FRL (FRLV, %) and the FRFx (FRFxV, %) was calculated using the following formula:

$$\mathrm{FRLV}(\mathrm{\%}) =\mathrm{ FRL }/ (\mathrm{TLV}-\mathrm{TV})\times 100$$

$$\mathrm{FRFxV}(\mathrm{\%}) =\mathrm{ FRFx }/ (\mathrm{TLV}-\mathrm{TV})\times 100$$

The increase of the FRL and FRFx were calculated using the following formula:

$$\mathrm{iFRL\%}= (\mathrm{FRLpost}-\mathrm{FRL})/\mathrm{FRL x }100$$

$$\mathrm{iFRFx\%}= (\mathrm{FRFxpost}-\mathrm{FRFx})/\mathrm{FRFx x }100$$

Additionally, for each patients we considered:

The BSA was calculated with the following formula:

$$\mathrm{BSA}= \surd \mathrm{ height}(\mathrm{cm})\mathrm{ x weight}(\mathrm{kg})/3600$$

  • The kinetic growth rate (KGR) was defined as the percentage-point difference between the liver volume and remnant liver before and after the intervention or surgery and it was calculated as percentage growth per day [10].

Outcome measures

Primary endpoint was to correlate the volumetric values with clinical outcomes, analysing the Anatomic Future Remnant Liver (FRL) and the the Functional Future Liver Remanent (FRFx) in patients with PHLF B,C and in patients with PHLF A.

Secondary end point was correlate Anatomic Future Remnant Liver (FRL) and the The Functional Future Liver Remanent (FRFx) in patients treated with ALLPS and with PVE followed by mayor hepatectomy.

Statistical analysis

Continuous variables reported were the median with interquartile range and categorical variables as frequency and absolute percentages. The normality criteria were tested on the cohort according to the Shapiro–Wilk test. The variables are compared with the Mann Whitney U test and chi-square for quantitative and qualitative data, respectively. Correlation between PHLF and post hepatic volume values were analysed with Spearman correlation.

For the statistical analysis, SPSS software (version 10, IBM SPSS, Chicago, IL, USA) was used and all tests were considered statistically significant with a value of p ≤ 0.05.

Source link