1. Introduction
Several recent studies have shown that parenchymal-sparing hepatic resections for malignant liver disease can increase the likelihood of liver salvage in case of recurrence, boosting the number of patients who would benefit from liver resection [
1]. These findings have led to an increased use of multiple nonanatomic resections (NAR) rather than major hepatectomies to achieve tumor clearance. However, parenchymal-sparing liver resection (PSR) requires a deep understanding of intraoperative navigation using ultrasonography (IOUS), particularly compared to simpler, conventional, major hepatectomies. During the latter, major vessels are identified and followed vertically, whereas during PSR, continuous employment of the IOUS is mandatory for identifying all terminal vascular branches that should be selectively resected or preserved [
2]. These new procedures rely on the evidence of communicating veins among adjacent main hepatic veins (HV) [
3,
4], which can be used to resect the proximal part of any major HV without affecting the outflow drainage of distal liver tissues. However, switching to intraoperative liver navigation with IOUS raises new issues in liver surgery: the preservation of vital liver tissues becomes fundamental but more difficult to achieve. These could lead to an increased rate of remnant liver ischemia (RLI) [
5], which can be caused either by remnant non-perfused liver tissue due to inappropriate inflow or by unintentional damage to tissue outflow. While inflow impairment has been thoroughly investigated, the changes occurring in the liver in cases of outflow deprivation have rarely been investigated, as at present, no available measurement of such an event has been reported. As a result, if the hepatic vein is occluded, what can be expected to happen to hepatic hemodynamics? This investigation explores outflow occlusion in a pig model using the study of the tissues’ absorbance in the visible and near-infrared spectra of light acquired by means of a hyperspectral camera.
2. Materials and Methods
This is a prospective observational non-randomized study in an animal model, conducted at the IRCAD Institute (Strasbourg, France) in October 2022. The aim of this study was to investigate the variations of four indexes (namely, oxygen saturation, StO2%; total hemoglobin index, THI; near-infrared perfusion index, NIR; total deoxy-hemoglobin, deoxy-Hb) among liver lobes with different outflow patency statuses, measured by means of a hyperspectral camera (details provided in the Methods section) at different time points. The considered time points were as follows: t0 = after laparotomy before hepatic vein occlusion; t1 = at the moment of hepatic vein occlusion; t2 = 10 min after occlusion; t3 = 30 min after occlusion. After t3, the vein was unclamped. T4 was 10 min after clamp removal.
To better understand the protocol, it can be useful to review the peculiarity of the anatomy of a pig’s liver. The liver is composed of five macroscopically independent lobes (right lateral lobe [RLL], right medial lobe [RML], left medial lobe [LML], left lateral lobe [LLL], and caudate lobe), each receiving its portal triads (inflow). Porcine hepatic veins are comprised of a left hepatic vein (LHV), a left middle hepatic vein (LMHV), a middle hepatic vein (MHV), a right middle hepatic vein (RMHV), and a right hepatic vein (RHV). The LHV, LMHV, and RHV drain the left lateral, left medial, and right lateral lobes, respectively. Only the right medial lobe is drained by two major hepatic veins, i.e., the MHV and RMHV. These two veins usually share a common trunk at the hepatocaval confluence. In order to conduct this experiment, the RML was purposely divided into two areas of vein drainage: the right side (RMLr) drained by the RMHV, and the left side (RMLl) drained by the MHV. No collateral veins between the hepatic veins have ever been described consistently with the macroscopic independence of the liver lobes. According to such peculiarities, the liver lobes investigated were the RLL (in which the outflow was totally occluded), the RML (both right and left sides, in which the RMLr drainage was occluded, and the RMLl was not), and the left lateral lobe (LLL, in which the outflow was always preserved and considered as the reference). Three different models were subsequently considered: the RLL model, in which the outflow was totally occluded without any possibility for alternative drainage; the RML model, in which the outflow was partially occluded and eventual interactions between regions drained by different main hepatic veins could occur; and, finally, the LLL model, in which the outflow was never interrupted and potential modifications could be related to intra-organ feedback.
2.1. Animals
The present study was part of the EXMachyna3 project (Intraoperative EXamination Using MAChine-learning-based HYperspectral for diagNosis & Autonomous Anatomy Assessment, Strasbourg, France), registered at ClinicalTrials.gov (NCT04589884) and approved by the local ethics committee of the School of Medicine at the University of Strasbourg (ID-RCB: 2020-A01896-33). All animals used in the experiment were managed in accordance with French regulations for animal use and care, and with the directives of the European Community Council (2010/63/EU) and ARRIVE guidelines [
6]. Six adult male swine (Sus scrofa ssp. domesticus, mean weight: 32.4 ± 4.4 kg) were housed and acclimatized for 48 h in an enriched environment, respecting circadian cycles of light–darkness, with constant humidity and temperature conditions. They fasted 24 h before surgery, with ad libitum access to water, and were finally sedated (zolazepam + tiletamine 10 mg/kg IM) 30 min before the procedure to decrease stress. Anesthesia was performed intravenously (18-gauge IV catheter in-ear vein) with Propofol 3 mg/kg and maintained with rocuronium 0.8 mg/kg along with inhaled isoflurane 2% via the automatic standard respiratory system. Vital parameters were monitored through a mechanical ventilator machine and heartbeat was monitored with a pulse oximeter (Mindray PM-60). At the end of the procedure, all animals were euthanized with a lethal dose of pentobarbital (40 mg/kg).
2.2. Surgical Procedure
A midline laparotomy was performed to access the abdominal cavity. Xiphoidectomies were routinely performed to maximize liver visibility. Intraoperative ultrasounds were routinely performed to study hepatic vein anatomy and to exclude anatomical variations that could impact the experimental results. In particular, in all pigs, the RHMV and the RHV had a common confluence at the vena cava, and the MHV had an independent insertion. The confluence of the RHV and the MHRV was dissected and isolated from right to left extrahepatically, and a vascular clamp was positioned at the time of clamping (t1). The efficacy of the clamping was confirmed by means of intraoperative Doppler ultrasound. After 30 min of venous clamping, the vascular clamp was removed. At each time point, capillary lactates were analyzed by puncturing Glisson’s capsule on the back side of the RLL and measured using a strip-based portable lactate analyzer, which presents a margin of error of 0.35 mmol/L (EDGE, ApexBio, Taipei, Taiwan).
2.3. Hyperspectral Imaging
A CMOS push-broom scanning hyperspectral camera (TIVITA, Diaspective Vision GmbH, Am Salzhaff, Germany) was used to generate HS images, performed with a camera-specific software module from the same company. It enables the acquisition of optical remission spectroscopy in the visible and near-infrared range (400–1000 nm) as a contact-free method without any influence on the measured tissue. The tridimensional hypercube is composed of a spatial resolution (x, y) plus a third dimension with the relative reflectance of each pixel (z). The range of the wavelength detected is 500–1000 nm with a 5 nm interval, totaling 100 wavelengths for every pixel. The scanning method is passed through a slit-shaped aperture, which is motorized with an internal stepper motor [
7]. The resolution of the hypercube is 640 × 480 pixels × 100 wavelengths. Image acquisition is performed at a distance of ~40 cm from the sample and monitored by a Bluefruit Feather nRF52832 distance sensor with an Adafruit VL53LOx device (Adafruit, New York, NY, USA) orthogonal to the liver’s surface. The light source is composed of six halogen lamps of 20 Watts (OSRAM Halospot 70, OSRAM GmbH, Munich, Germany). The HS camera takes about six seconds to perform the acquisition of the hypercube, which is transferred to a PC where it is processed, creating pseudo-color images. The relative reflectance:
(I/I0)
is converted into relative absorbance through the equation
A = −log(I/I0).
The device used in this experiment provides different algorithms (preset) that quantify the relative oxygen saturation (StO
2%) of microcirculation at a depth of ~1 mm, and at deeper layers with the near-infrared (NIR) spectrum (3–5 mm) [
8]. Briefly, StO
2% is calculated with an algorithm based on the second derivative of the absorption spectra (570–590 nm and 740–780 nm). The NIR perfusion index is calculated with the absorbance spectra in a spectral range of 655–735 nm and 825–925 nm.
Another algorithm allowed us to estimate the total hemoglobin index (THI), which represents the sum of the oxygenated and deoxygenated hemoglobin in the tissue explored. Relatively deoxygenated hemoglobin (deoxy-Hb) was estimated as a consequence of the formula that enabled estimating the StO
2% and the THI by subtracting the percentage of oxygenated hemoglobin from the THI value. Quantitative analysis of those four indexes was performed intraoperatively using the TIVITA Suite software module over four regions of interest (ROI), manually set in the RLL, RMLr, RMLr, and LLL. An example of the ROI positioning is depicted in
Figure 1. The methods and algorithms of the TIVITA system were explained in more detail by Holmer et al. in 2018 [
8].
2.4. Statistical Analysis
Data were reported as mean and standard deviation (SD). Normality was assessed with the Kolmogorov–Smirnov test at each time point. The variation of the four indexes among liver lobes was measured at each time point by fitting a linear mixed model (estimated using REML and nloptwrap optimizer). The models included the pigs’ IDs as a random effect (formula: ~1|ID). Standardized parameters were obtained by fitting the model on a standardized version of the dataset, and 95% confidence intervals (CIs) and p values were computed using a Wald t-distribution approximation. The same analysis was repeated after recording measurements in the LLL to account for the differences among the three occluded models. This latter analysis was repeated twice, first with RLL as the reference, and then with RMLl as a reference. All tests were two-tailed, and a p value < 0.05 was considered significant. The analyses were performed using R software (v. 4.2.2, libraries: lme4, ggplot2, report, pavo).
3. Results
An example of the photos acquired with the TIVITA system at each time point is depicted in
Figure 2, showing the visible change on the liver’s surface after partial outflow occlusion.
The measured absorbance within the spectra among lobes at each time point was depicted in
Supplementary Figure S1A–E. The trends of the indexes measured with the hyperspectral camera and the lactate are reported in
Figure 3A–E at each time point. Each index was then analyzed to estimate its variation among the different lobes per each time point. Considering the lactate variation over time, no significant differences were measured, with an almost linear trend over time (beta = −0.04, 95% CI [−0.24, 0.16], t(96) = −0.39,
p = 0.698; Std. beta = −0.03, 95% CI [−0.17, 0.11]).
3.1. StO2% Variations
The model’s total explanatory power for StO2% at t0 is substantial (conditional R2 = 0.85), and the part related to the fixed effects alone (marginal R2) is 9.37 × 10−3. The model’s intercept, corresponding to site = LLL, is at 38.33 (95% CI [22.61, 54.06], t(18) = 5.12, p < 0.001). No significant differences were observed between lobes.
At t1, the model’s total explanatory power is substantial (conditional R2 = 0.82), and the part related to the fixed effects alone (marginal R2) is 0.76. The model’s intercept is at 50.80 (95% CI [42.96, 58.64]). Within this model: (1) the effect of the site [RLL] is statistically significant and negative (beta = −31.40, 95% CI [−41.05, −21.75], t(14) = −6.98, p < 0.001, Std. beta = −1.93); (2) the effect of the site [RMLr] is statistically significant and negative (beta = −24.40, 95% CI [−34.05, −14.75], t(14) = −5.42, p < 0.001, Std. beta = −1.50).
At t2, the model’s total explanatory power is substantial (conditional R2 = 0.79), and the part related to the fixed effects alone (marginal R2) is 0.37. The model’s intercept is at 51.00 (95% CI [36.28, 65.72]). Within this model: (1) the effect of the site [RLL] is statistically significant and negative (beta = −30.67, 95% CI [−42.69, −18.65], t(18) = −5.36, p < 0.001, Std. beta = −1.49); (2) the effect of the site [RMLr] is statistically significant and negative (beta = −21.83, 95% CI [−33.85, −9.81], t(18) = −3.82, p = 0.001, Std. beta = −1.06). At t3, the model’s total explanatory power is substantial (conditional R2 = 0.84), and the part related to the fixed effects alone (marginal R2) is 0.52. The model’s intercept is at 56.02 (95% CI [41.17, 70.87]). Within this model: (1) the effect of the site [RLL] is statistically significant and negative (beta = −31.42, 95% CI [−43.86, −18.98], t(13) = −5.46, p < 0.001, Std. beta = −1.55); (2) the effect of the site [RMLr] is statistically significant and negative (beta = −22.42, 95% CI [−34.86, −9.98], t(13) = −3.89, p = 0.002, Std. beta = −1.11). At t4, the model’s total explanatory
power is substantial (conditional R2 = 0.79), and the part related to the fixed effects alone (marginal R2) is 0.05. The model’s intercept is at 57.25 (95% CI [39.54, 74.96]); no significant variations among lobes were identified. Results are summarized in
Table 1 (random effects results in
Supplementary Table S1).
3.2. Deoxy-Hb Variations
At t0, the model’s total explanatory power is substantial (conditional R2 = 0.77), and the part
related to the fixed effects alone (marginal R2) is 8.92 × 10−3. The model’s intercept,
corresponding to site = LLL, is at 40.87 (95% CI [27.03, 54.71], t(18) = 6.20, p < 0.001). No significant variations were observed among lobes at this time point.
At t1, the model’s total explanatory power is substantial (conditional R2 = 0.77), and the part related to the fixed effects alone (marginal R2) is 8.92e−03. The model’s intercept is at 40.87 (95% CI [27.03, 54.71]). Within this model, none of the lobes reached significance. At t2, the model’s total explanatory power is substantial (conditional R2 = 0.92), and the part related to the fixed effects alone (marginal R2) is 0.40. The model’s intercept is at 26.72 (95% CI [16.08, 37.37]). Within this model: (1) the effect of the site [RLL] is statistically significant and positive (beta = 24.14, 95% CI [18.73, 29.55], t(18) = 9.37,
p < 0.001, Std. beta = 1.57); (2) the effect of the site [RMLr] is statistically significant and positive (beta = 10.35, 95% CI [4.94, 15.76], t(18) = 4.02,
p < 0.001,Std. beta = 0.67). At t3, the model’s total explanatory power is substantial (conditional R2 = 0.81), and the part related to the fixed effects alone (marginal R2) is 0.40. The model’s intercept is at 21.11 (95% CI [7.03, 35.19]). Within this model: (1) the effect of the site [RLL] is statistically significant and positive (beta = 27.62, 95% CI [16.47, 38.76], t(14) = 5.32,
p < 0.001, Std. beta = 1.53); (2) the effect of the site [RMLr] is statistically significant and positive (beta = 17.00, 95% CI [5.86, 28.15], t(14) = 3.27,
p = 0.006, Std. beta = 0.94). At t4, the model’s total explanatory power is substantial (conditional R2 = 0.81), and the part related to the fixed effects alone (marginal R2) is 0.07. The model’s intercept is at 24.81 (95% CI [9.31, 40.32]), and no significant variations among lobes were detected. Results are summarized in
Table 1 (random effects results in
Supplementary Table S1). Temporal trends were depicted in
Figure 3B. Comparisons among occluded lobes after the exclusion of the LLL were reported in
Table 2 (random effects results in
Supplementary Table S2) when compared to RLL, and in
Supplementary Table S3 when compared to RMLl.
3.3. NIR Variations
At t0, the model’s total explanatory power is substantial (conditional R2 = 0.32), and the part
related to the fixed effects alone (marginal R2) is 3.52e−03. The model’s intercept,
corresponding to site = LLL, is at 58.33 (95% CI [47.56, 69.11], t(18) = 11.37, p < 0.001). No significant variations were observed among lobes at this time point. At t1, the model’s total explanatory power is substantial (conditional R2 = 0.35), and the part related to the fixed effects alone (marginal R2) is 0.35. The model’s intercept is at 53.60 (95% CI [40.10, 67.10]). Within this model, the effect of the site [RLL] is statistically significant and negative (beta = −25.00, 95% CI [−44.01, −5.99], t(14) = −2.82, p = 0.014, Std. beta = −1.51).
At t2, the model’s total explanatory power is substantial (conditional R2 = 0.35), and the part related to the fixed effects alone (marginal R2) is 0.35. The model’s intercept is at 53.60 (95% CI [40.10, 67.10]). Within this model, the effect of the site [RLL] is statistically significant and negative (beta = −25.00, 95% CI [−44.01, −5.99], t(14) = −2.82,
p = 0.014, Std. beta = −1.51). At t3, the model’s total explanatory power is substantial (conditional R2 = 0.33), and the part related to the fixed effects alone (marginal R2) is 0.13. The model’s intercept is at 52.44 (95% CI [35.10, 69.78]). Within this model, no significant variations between lobes were observed. At t4, the model’s total explanatory power is substantial (conditional R2 = 0.80), and the part related to the fixed effects alone (marginal R2) is 0.09. The model’s intercept is at 65.75 (95% CI [55.87, 75.63]). Within this model, the effect of the site [RLL] is statistically significant and negative (beta = −6.75, 95% CI [−13.30, −0.20], t(10) = −2.30,
p = 0.045, Std. beta = −0.80). Results are summarized in
Table 1 (random effects results in
Supplementary Table S1). Temporal trends were depicted in
Figure 3C. Comparisons among occluded lobes after the exclusion of the LLL were reported in
Table 2 (random effects results in
Supplementary Table S2) when compared to RLL, and in
Supplementary Table S3 for RMLl.
3.4. THI Variations
At t0, the model’s total explanatory power is substantial (conditional R2 = 0.59), and the part
related to the fixed effects alone (marginal R2) is 0.03. The model’s intercept,
corresponding to site = LLL, is at 65.17 (95% CI [54.64, 75.69], t(18) = 13.01, p < 0.001). No significant variations were observed among lobes at this time point.
At t1, the model’s total explanatory power is substantial (conditional R2 = 0.59), and the part related to the fixed effects alone (marginal R2) is 0.03. The model’s intercept is at 65.17 (95% CI [54.64, 75.69]). Within this model, the variations in the index between lobes were not significant. At t2, the model’s total explanatory power is substantial (conditional R2 = 0.66), and the part related to the fixed effects alone (marginal R2) is 0.16. The model’s intercept is at 52.83 (95% CI [42.56, 63.11]). Within this model, the effect of the site [RLL] is statistically significant and positive (beta = 10.67, 95% CI [1.38, 19.95], t(18) = 2.41,
p = 0.027, Std. beta = 0.86). At t3, the model’s total explanatory power is substantial (conditional R2 = 0.77), and the part related to the fixed effects alone (marginal R2) is 0.15. The model’s intercept is at 55.54 (95% CI [43.68, 67.41]). At t4, the model’s total explanatory power is substantial (conditional R2 = 0.74), and the part related to the fixed effects alone (marginal R2) is 0.14. The model’s intercept is at 53.75 (95% CI [41.08, 66.42]), but no significance has been detected. Results are summarized in
Table 1 (random effects results in
Supplementary Table S1). Temporal trends are depicted in
Figure 3D. Comparisons among occluded lobes after the exclusion of the LLL are reported in
Table 2 (random effects results in
Supplementary Table S2) when compared to RLL, and in
Supplementary Table S3 when compared to RMLl.
4. Discussion
The lobes impacted by HV occlusion showed significant variations in the hemodynamics parameter measured by our study when compared to those with preserved outflow. In particular, StO2% decreased significantly at each time point after occlusion in the RLL and in the RMLr when compared to LLL, returning to normal after outflow restoration, while deoxy-Hb showed the opposite behavior, with a significant increase over time. The NIR index, a parameter estimating perfusion deeper than StO2%, decreased after venous occlusion in the RLL and RMLr, but remained significantly decreased only in RLL at the other time points, returning to normal after declamping. These variations were probably the direct indication of liver congestion; when outflow is occluded, the amount of deoxygenated hemoglobin increased, as a consequence of the impossibility of effectively draining venous blood.
It is of interest that the lactates did not vary during time points, and they were not related to the hemodynamic indexes; although this parameter is the most widely used in the liver surgical field to monitor circulatory impairment, with an evident predictor role in case of inflow occlusion [
9], they seemed to be useful in cases of outflow occlusion. Consequently, their role in estimating venous congestion was not validated by our measurements, demonstrating that the hyperspectral indexes can provide more robust evidence of the changes occurring in liver tissue.
When considering the temporal trends, other crucial information can be derived from our study. The indexes in the LLL were always very stable over time, and a similar and superimposable behavior was also recognized on the left side of the RML, where outflow was ensured by the MHV. Notwithstanding, the counterpart of the RML, the right side, showed trends closer to the totally excluded lobe (the RLL). The RML, although fed by two hepatic veins, is macroscopically a single structure, where indeed connections between veins were not recorded on the IOUS before and during clamping. However, the variations of the indexes over time were less sharp than those observed in the RLL. When compared to the variations in RMLl, a more contained reduction in deoxy-Hb and StO
2% was observed compared to that observed in the RLL. As a confirmation, when the RMLl was taken as the reference, in RMLr, StO
2% significantly increased and deoxy-Hb decreased over time during the occlusion period compared to the trends in the RLL. This observation can be attributed to the presence of venous communications among the clamped RMHV and the MHV that, even if not visualized on the IOUS, became patent after outflow occlusion to create a sort of compensation for the outflow pathway. This phenomenon was probably not observed in the RLL due to the anatomic separation between the RLL and the RML. Recently, the presence and the role of communicating veins (CV) between adjacent HVs were described in surgical human settings [
10,
11]; these veins were recognized in up to 80% of livers intraoperatively, even if they could not always be visualized directly via IOUS due to their small size [
4]. More importantly, in autopsy specimens, it has been demonstrated that these veins were present in almost all cases [
12]. It has also been reported that, when CVs are not visualizable, flow assessment using color Doppler US or e-flow US can confirm their presence by the absence of portal flow inversion (from hepatopetal to hepatofugal) [
2]. Again, this change of direction in the portal flow is a rare finding. The physiological changes behind flow inversion rely on the evidence that, when the outflow is occluded, the portal vein seems to become the outflow pathway in the absence of CVs, while the arterial flow is increased, becoming the main inflow route [
13,
14]. In cases when no CVs are present, in our RML model, one can expect to observe different variations in the hemodynamic indexes of the RMLr, with rebound feedback in the RMLl. However, our measurements showed that the behavior of the RMLl was almost stable over time, exactly as the LLL, which was never occluded. Indeed, index variations were observed only in the RMLr, where the reduction of oxygenated blood and the increase of deoxygenated blood were slightly less significant when compared to the total excluded lobe, and, in general, more rapid in restoring the original physiological condition. We hypothesized that these findings suggested the presence and the opening of CVs among the clamped RMHV and the patent MHV. Accordingly, hyperspectral estimation of those indexes could be a novel method to estimate the presence of CVs in an intraoperative setting. However, this conclusion should be carefully considered, as it is mostly speculative, and other studies in human settings during clamping using color Doppler or e-flow IOUS should still be performed.
Potential surgical applications of the hyperspectral camera’s indexes have already been investigated [
15,
16,
17], but another valuable application of our method could be the intraoperative estimation of remnant liver congestion. Recently, it has been reported that remnant liver ischemia should be carefully considered in human oncological settings because of a significantly increased risk of morbidity, recurrence, and mortality [
5]. Such a phenomenon may occur not only in cases of altered portal vascularization of a liver segment but also in cases of preservation of liver tissue without appropriate outflow drainage. At present, there are no means to estimate appropriate outflow preservation of the remnant liver intraoperatively, and the finding is always recognized during postoperative CT scans. In our totally occluded model (the RLL), the trends among the times of StO
2% and deoxy-Hb showed a clear trend that could be applied in operating rooms to estimate congestion. Consequently, our method could provide comparative assessment between different sites of the remnant liver, as well as with the liver prior to transection. This application can be performed in real time, as it is not time-consuming. Moreover, if our measurements are confirmed, the hyperspectral camera could be considered to estimate intraoperatively the remnant liver ischemia (RLI) [
5], which has been recently associated with an increased rate of postoperative complications and worse long-term results.
The present study has several limitations. Fundamentally, it is an observational study, without any comparison of clinical effects. This means that the clinical implications put forward should be carefully evaluated in further studies, and they should be considered speculative at this time. In addition, remission spectrometry in in vivo tissues cannot provide a real quantification of the concentration of hemoglobin as expected; e.g., in the case of a biochemical analysis. This means that the values presented here (a part of SatO2%) are absolute values without a specific unit of measure. Another limitation is that, regarding the use of the indexes as markers of liver venous congestion, no cut-offs were provided in this study because of its design; even in this case, other studies should be conducted to further clarify the details of the potential clinical applications.