2.1.

Vis-DRS Instrument

The setup, as shown in Fig. 1(a), consists of a self-calibrating fiber optic surface probe and a portable vis-DRS system equipped with a laptop computer for data acquisition.³³ Our spectroscopy system includes a white-light emitting diode (LED) as the light source and two visible USB-spectrometers (Avantes BV, The Netherland) for spectrum detection. The fiber optic probe is composed of nine $200/220\text{-}\mu \mathrm{m}$ multimode optical fibers, with six for DRS illumination, the seventh one for DRS detection, and the remaining two for self-calibration illumination and detection, respectively. Seven illumination fibers are the highest number that can be included inside a standard 3-mm fiber optic jacket. At the distal end, the six DRS illumination fibers are arranged as a ring around the DRS detection fiber [Fig. 1(b)]. The source-detection separation (SDS), which is the center-to-center distance between the DRS illumination ring and the detection fiber, is 1.2 mm. The SDS is a tradeoff between a large penetration depth and good signal-to-noise ratio in liver tissue. The self-calibration and detections fibers are terminated inside the rigid part of the probe using a diffusely reflective coating material (not shown in Fig. 1).³³ All illumination fibers are connected to the white LED, and the detection fibers are connected to the two spectrometers, respectively. The real-time diffuse reflectance and self-calibration spectra (430 to 630 nm) are collected by a custom LabVIEW program. For calibration, the tissue spectrum is divided by the self-calibration spectrum collected concurrently to account for real-time instrument drifts and fiber bending loss.³⁴ A 3D-printed dark stopper is added to the distal end of the probe to maintain a stable tissue contact while blocking background noise from the room light.

Fig. 1

(a) Photograph of the visible diffuse reflectance spectroscopy (vis-DRS) instrument and self-calibration fiber optic surface probe used for the study and (b) diagram of the fiber layout at the distal end of the probe. The center-to-center distance between the illumination ring and detection fiber is 1.2 mm.

2.2.

Phantom Validation

Tissue mimicking liquid phantoms were used to evaluate the performance of the instrument in measuring optical properties of highly absorptive liver tissue. The phantoms were created using human Hb powder (H0267, Sigma-Aldrich Co. LLC) as the absorber and $1\text{-}\mu \mathrm{m}$ polystyrene microspheres (07310-15, Polysciences Inc) as the scatterer. The phantoms were obtained through 16 successive titrations of the Hb from 46.36 to $112.01\mu \mathrm{M}$ and fixing the number of scatterers. The absorption coefficients (${\mu }_a$) of the phantoms were independently determined by measuring the ${\mu }_a$ of the Hb stock solution using a spectrophotometer (Lambda35, PerkinElmer) and the Beer–Lambert law. The reduced scattering coefficients (${{\mu }_s}^{\prime }$) were calculated using Mie theory with known particle refractive index, size, and density. Diffuse reflectance and self-calibration spectra in the wavelength range of 435 to 630 nm for each of the 16 phantoms were collected using the fiberoptic probe and spectrometer. Self-calibration was performed by dividing the phantom spectrum by the calibration spectrum (point-by-point). A Monte Carlo inverse model was used to extract the phantom ${\mu }_a$ and ${{\mu }_s}^{\prime }$ and Hb concentration from the calibrated phantom spectrum, following the procedures described by Yu et al.³⁴ Finally, the percent errors, which are the difference between the extracted and expected mean values in ${\mu }_a$ and ${{\mu }_s}^{\prime }$, were computed.

2.3.

Basic Surgical Preparation

The use of animals in this study was approved by the Institutional Animal Care and Use Committees (IACUC) at the Medical College of Wisconsin (MCW) and Zablocki Veterans Affairs Medical Center. The animal use programs at both institutions are fully accredited by the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC) International. Experiments were carried out on adult (42 to 52 kg) female pigs. The animals fasted for 12 h and were premedicated/sedated with an intramuscular (IM) injection of Tiletamine 3 to $5\mathrm{mg}/\mathrm{kg}$, Zolazepam 3 to $5\mathrm{mg}/\mathrm{kg}$, and xylazine $2.2\mathrm{mg}/\mathrm{kg}$. When the animal was sedated enough, it was taken into the intubation suite. Atropine IM ($50\mathrm{mcg}/\mathrm{kg}$) was given, and an auricular intravenous (IV) catheter was placed prior to intubation. Ocular reflexes were tested for depth of anesthesia, and additional propofol (1 to $2\mathrm{mg}/\mathrm{kg}$) was administered to achieve appropriate intubating conditions. A single dose of enrofloxacin ($5\mathrm{mg}/\mathrm{kg}$) IM was administered as antibiotic prophylaxis because of the long, invasive surgical preparation. The animal was transferred intubated and ventilated under continuous pulse oximetry to the surgical suite, where they were continuously ventilated using an anesthesia machine (Ohmeda CD, GE, Datex Ohmeda, Madison, Wisconsin). General anesthesia was maintained with 1.5% to 3% isoflurane. Inspiratory oxygen fraction, expiratory carbon dioxide concentration, and expiratory isoflurane concentration were continuously displayed with an infrared analyzer (POET II, Criticare Systems, Waukesha, Wisconsin). An esophageal probe was placed for non-invasive cardiac output (CO) measurements in the initial experiments, where CO values values showed good correlation with blood pressure and pulse pressure amplitude (ECOM, ECOM Medical Inc, San Juan Capistrano, California). The animal was placed in the supine position. The right groin was infiltrated with 1% lidocaine, and femoral venous and arterial catheters were placed through a cut down technique for monitoring, drug administration, and blood sampling. The midline of the abdomen was infiltrated with 1% lidocaine with extension to the subcostal areas bilateral and the sternum before incision. Throughout the experiments, care was taken to increase anesthetic depth for any signs of light anesthesia, e.g., an increase in blood pressure or heart rate. The animal was maintained at $38.5\pm 0.5°\mathrm{C}$ with a warming blanket.

2.4.

Statistical Analysis

The vis-DRS values for tissue scattering and absorption were derived using the Monte Carlo inverse model. A sub-analysis to determine whether the inverse model needed correction for the presence of bile in the liver tissue quantified the extent of agreement between the measurements with and without correction for biliverdin using Lin’s concordance correlation coefficient with standard error adjusted for within-animal replicate measures.^35,36 Blood gas values were compared with vis-DRS values calculated with and without correction for biliverdin using the overlapping dependent correlations test of Meng et al.³⁷ Analysis showed that correction for biliverdin absorption did not change the values for scattering and absorption in the (physiologic) Hb range of interest, and no correction for biliverdin was used for the other analyses in this study.

Post-hoc data reduction, data plotting and statistical analysis of the pooled data were performed using SigmaPlot 11 (Systat Software, Richmond, California) and R 3.5.0 (R foundation for statistical Computing, Vienna, Austria). Values are expressed as $\text{mean}\pm \mathrm{SE}$ or median/ range as appropriate. Statistical significance was assumed for $p<0.05$. Corrections for multiple comparisons were made within each analysis. The statistical tests for each sub-study are described in the results section.

3.1.

Phantom Validation

The extracted mean ${\mu }_a$ and ${{\mu }_s}^{\prime }$ values of the 16 phantoms are plotted against their expected values in Fig. 2. The extracted mean values and error bars were obtained by averaging over ${\mu }_a$ and ${{\mu }_s}^{\prime }$ values (wavelength averaged) extracted using each of the 16 phantoms as a reference and all phantoms as targets.³⁴ The mean ${\mu }_a$ and ${{\mu }_s}^{\prime }$ of the 16 phantoms ranged from 3.87 to $9.35{\mathrm{cm}}^{-1}$, and 17.33 to $5.96{\mathrm{cm}}^{-1}$, respectively. Average errors of 1.86% and 4.43% were obtained for measuring ${\mu }_a$ and ${{\mu }_s}^{\prime }$, respectively.

Fig. 2

Extracted versus expected phantom mean values for (a) absorption ${\mu }_a$ and (b) scattering ${{\mu }_s}^{\prime }$ from 16 phantom experiments. Mean ${\mu }_a$ phantoms range: 3.87 to $9.35{\mathrm{cm}}^{-1}$ with average error of 1.86%. Mean ${{\mu }_s}^{\prime }$ phantoms range: 17.33 to $5.96{\mathrm{cm}}^{-1}$ with average error of 4.43%.

3.2.

Data Variance (in Vivo and ex Vivo)

Twenty data sets were obtained in three animals using the in vivo and ex vivo preparation. Each data set consisted of 20 measurements in a 2-min time frame. Data sets were obtained during steady-state conditions and included the whole spectrum of clinically relevant hepatic tissue saturations (29% to 88% for the specific analysis). The coefficient of variation for each data set ranged from 0.7% to 1.6% of the mean value, suggesting that measurement errors were small and independent of the experimental conditions, i.e., Hb concentrations, ${\mathrm{PaO}}_2$, hepatic blood flow, or perfusion pressure.

3.3.

Effects of Correction for the Absorber Biliverdin on Scattering and Absorption

Seventy-five data points were obtained from the cardiopulmonary bypass protocol (ex vivo) (2.2.2). In general, values for absorption and scattering with and without biliverdin correction correlated closely between analyses. Lin’s concordance correlation coefficients (95% confidence limits) were 0.991 (0.987, 0.994) for absorption and 0.959 (0.943, 0.971) for scattering. The difference between values for absorption and scatter when biliverdin was and was not included as an absorber in the spectral analysis was small at high values and larger at low values (for ${\mu }_a$: $-0.058\pm 0.02{\mathrm{cm}}^{-1}$, $p=0.009$, for ${{\mu }_s}^{\prime }$: $-0.12\pm 0.02{\mathrm{cm}}^{-1}$, $p=0.0002$) [Fig. 3(a)]. For scattering, the difference between values was greater at lower Hb values ($-0.051\pm 0.03{\mathrm{cm}}^{-1}$, $p=0.047$).

Fig. 3

Effects of including biliverdin as an absorption factor in the inverse Monte Carlo model. (a) Values for absorption (${\mu }_a$, left) and scattering (${{\mu }_s}^{\prime }$, right). Values were obtained from the perfusion of liver with cardiopulmonary bypass (ex vivo) protocol. Data points were color-coded for the Hb levels at the time of sampling. Correction for biliverdin resulted in slightly higher ${{\mu }_s}^{\prime }$ values; however, this was not significantly different. Thirty-three data points. (b) Correlation of Hb measured by vis-DRS and blood gas machine (left) with correction for biliverdin and (right) without correction for biliverdin. About 33 data points from three different animals. Data points were color-coded for inspiratory oxygen fraction (${\mathrm{FiO}}_2$) at the time of sampling. Data were log-transformed to allow for linear regression analysis. Correction for biliverdin did not significantly affect the Hb values.

3.4.

Correlation between Blood Gas Analyzer Hemoglobin and vis-DRS Hemoglobin (in Vivo Hemodilution)

To evaluate the correlation between the Hb values determined with the vis-DRS method and the values obtained by blood gas analysis, 33 data points were obtained from three animals that underwent the in vivo hemodilution protocol to extreme anemia. As shown in Fig. 3(b), there was a close correlation between Hb values measured with the blood gas machine and the values measured with the vis-DRS probe, independent of whether values were corrected for absorption by biliverdin ($R^2=0.81$) or not ($R^2=0.85$). There was no significant difference between two correlation coefficients ($p=0.38$), and vis-DRS values with and without correction for biliverdin correlated closely ($R^2=0.9$). Due to the excellent correlation between the values for absorption, scattering (3.3.), and Hb (3.4.), we did not include biliverdin as a separate absorber in the Monte Carlo inversion model for the following analyses.

3.5.

Correlation between Blood Gas Analyzer SO₂ and vis-DRS SO₂ (ex Vivo)

We used 39 data points from two pig livers perfused via cardio-pulmonary bypass. Values for ${\mathrm{PO}}_2$ and calculated ${\mathrm{SO}}_2$ obtained from the blood gas samples followed the curve obtained with the Hill equation for pigs with a Hill coefficient of 3.02 and $P_{50}$ of 32.9 mmHg [Fig. 4(a)].³⁸

Fig. 4

Finding the best oxygen partial pressure (${\mathrm{PO}}_2$)—oxygen saturation (${\mathrm{SO}}_2$) curve fit for vis-DRS data. (a) Theoretical curve fit: ${\mathrm{SO}}_2$ was calculated from the ${\mathrm{PO}}_2$ measured with the blood gas machine using the theoretical Hill coefficient of 3.02 and ${\mathrm{PO}}_2$ 32.9 mmHg. Goodness-of-fit confirmed excellent fit with the expected Hb-oxygen dissociation curve (black line).³⁸ For pooled data: $R^2=0.96$ for pH 7.3 to 7.4: $R^2=98.1\%$; for pH 7.1 to 7.25: $R^2=94.4\%$. (b) Actual curve: vis-DRS values plotted against the same ${\mathrm{PO}}_2$ values. The Vis-DRS measurements did not follow the theoretical ${\mathrm{PO}}_2/{\mathrm{SO}}_2$ curve well but showed a meaningful correlation. (c) Obtaining the best customized curve fit: the vis-DRS data for both pH ranges were pooled. After testing multiple models with varying Hill coefficients and $P_{50}$, we found a good fit with a Hill coefficient of 1.67 and $P_{50}=34\mathrm{mmHg}$. The goodness-of-fit was $R^2=0.81$. Data points and fitted curves are color-coded for pH; size indicates Hb levels at the time of sampling. Thirty-nine data points.

The resulting ${\mathrm{PO}}_2$-vis-DRS curves did not match the theoretical curve; however, the custom-fitted Hill plot showed a meaningful correlation. Three models were evaluated. Model 1 allowed $P_{50}$ values to vary by pH condition with the Hill coefficient fixed at its value estimated based on the custom fit. This resulted in $P_{50}$ (pH 7.1 to 7.25) = 40.8 mmHg, $P_{50}$ (pH 7.3 to 7.4) = 32.7 mmHg, and Hill coefficient: 2.68. Model 2 allowed both the $P_{50}$ values and the Hill coefficient to vary by pH condition. For pH 7.1 to 7.25, this resulted in $P_{50}=40.9\mathrm{mmHg}$ and a Hill $\text{coefficient}=2.53$ and for pH 7.3 to 7.4 in $P_{50}=32.8\mathrm{mmHg}$ and a Hill $\text{coefficient}=2.88$. This model had a goodness-of fit of $R^2=0.96$ for both datasets. Model 3 used the same $P_{50}$ value and Hill coefficient for both pH conditions, resulting in a $P_{50}=34\pm 1.2\mathrm{mmHg}$ and Hill $\text{coefficient}=1.67\pm 0.15$ This model had a goodness-of-fit of $R^2=0.81$, which was not significantly different from model 2 ($p=0.434$) but provided a better fit than the theoretical Hill coefficient ($p<0.001$). Since in (clinical) practice pH is not continuously measured but generally varies within the range of 7.1 to 7.4, we decided to use the coefficients derived from pooled data for further analysis [Fig. 4(c)].

The results suggested that vis-DRS measurements accurately reflected changes in tissue ${\mathrm{PO}}_2$ but that a mathematical transform had to be added to the Monte-Carlo inverse model to reflect the true ${\mathrm{SO}}_2$ value. Since both ${\mathrm{SO}}_2$ and vis-DRS followed the Hill curve, their logit-transformed versions were linearly related [Fig. 5(a)]. This allowed fitting of a conversion model

Fig. 5

Creating a conversion model for vis-DRS data. (a) ${\mathrm{SO}}_2$ and vis-DRS data were logit-transformed to achieve linearity. This allowed the creation of a conversion formula to match the vis-DRS data with the expected ${\mathrm{SO}}_2$ values. (see Sec. 3.5). (b) Converted vis-DRS data were plotted against theoretical ${\mathrm{SO}}_2$ values. Goodness-of-fit for the pooled data were $R^2=0.76$. The data are color-coded for pH; however, the transform was created from the pooled values for both pH levels.

We developed an adjustment formula by solving the theoretical Hill equation and the Hill equation fitted to the vis-DRS data or ${\mathrm{pO}}_2$. Equating the two equations allowed solving for ${\mathrm{SO}}_2$

3.6.

In Vivo Vascular Occlusion

We investigated the ability of the vis-DRS method to reliably identify changes in tissue saturation during total occlusion of the hepatic vessels. The analysis included 21 data-points from two animals [Fig. 6(a)].

Fig. 6

(a) Tissue oxygen saturation as measured by vis-DRS during occlusion of the hepatic artery and/or portal vein. Hepatic artery occlusion: ${\mathrm{\Delta }}_{\mathrm{vis}-\mathrm{DRS}}$: $-44.4\pm 6.9\%$, $p<0.001$. Portal vein occlusion: ${\mathrm{\Delta }}_{\mathrm{vis}-\mathrm{DRS}}$: $-54.3\pm 6.2\%$, $p<0.001$. A total of 21 data points were used from two animals. (b) Selective right hepatic artery occlusion shows selective tissue desaturation in the right liver lobe. ${\mathrm{\Delta }}_{\mathrm{vis}-\mathrm{DRS}}-14.7\pm 1.39\%$, $p=0.0018$. Desaturation promptly recovered when the vessel was reopened. There was no decrease in tissue saturation in the left liver lobe.

Vis-DRS was able to identify a decrease in tissue oxygen saturation by $44.4\pm 6.9\%$ from baseline with occlusion of the hepatic artery, which contributes the smaller part of the total oxygen delivery to the liver ($p<0.001$). Occlusion of the portal vein, which contributes $\sim 70\%$ of the total oxygen delivery to the liver, reduced vis-DRS by $54.3\pm 6.2\%$ ($p<0.001$).

To confirm selectivity of this effect to a vascular bed, we selectively occluded the right hepatic artery while the left remained patent. Tissue saturation in the perfusion area of the right hepatic artery (right liver lobe) dropped by $14.7\pm 1.4\%$ ($p=0.0018$), while saturations slightly increased in the perfused area (left lobe) area (${\mathrm{\Delta }}_{\mathrm{vis}-\mathrm{DRS}}$: $4.8\pm 1.4\%$, $p=0.04$, linear regression model). Saturations returned to baseline after reopening the vessel [Fig. 6(b)].

4.1.

Use of Visible Diffuse Reflectance Spectroscopy (vis-DRS) in Low Absorptive Organs in Vivo

Vis-DRS reflected changes in myocardial tissue saturation during coronary occlusion in an in vivo pig model.³² Cardiac tissue saturation was consistently $\sim 5\%$ higher than coronary venous saturation, which was derived via blood gas sampling. Both values changed similarly with different conditions, suggesting that the tissue saturation measured by the probe correlated with the blood oxygen saturation.³² A study using vis-DRS to determine myocardial tissue oxygenation before and after coronary artery bypass surgery in human patients demonstrated wider data variance, which pointed to possible limitations in the interpretation of individual values.³⁹ Tissue saturation values could not be validated due to the inability to obtain coronary venous blood samples in human patients. In general, tissue saturations in areas distal to a stenotic coronary artery increased with improved perfusion after bypass surgery. However, baseline tissue saturations proximal to the stenosis were often lower than those in the supposedly ischemic area distal to the stenosis. Vis-DRS values for cardiac tissue in human patients³⁹ were substantially lower compared with those measured in healthy pigs.³² It is not clear whether this indicated decreased oxygen delivery in hearts with ischemic vascular disease⁴⁰ or differences due to the individual technology.

Vis-DRS has also been used to determine tissue saturation in gastrointestinal mucosa³⁰ and muscle in pigs and gastrointestinal mucosa, buccal mucosa, facial skin, and finger muscle in human volunteers and patients.³¹ Muscle tissue saturation in pigs was between 20% and 35% lower than arterial oxygen saturation determined with pulse oximetry. Decreasing ${\mathrm{FiO}}_2$ decreased both saturations in a linear fashion and with excellent correlation ($R^2=0.98$); however, vis-DRS values decreased more than pulse oximetry. Similarly, vis-DRS values decreased substantially more than pulse oximetry during forefinger ischemia and global ischemia in human subjects. This suggests that pulse oximetry may not be a good method to validate vis-DRS values, in particular, since pulse oximetry overestimates ${\mathrm{SaO}}_2$ at values $<85\%$ and becomes highly unreliable for ${\mathrm{SaO}}_2<66\%$.^40,41

4.2.

Effect of Tissue Homogeneity on vis-DRS Values

Discrepancies between vis-DRS values and tissue saturation values obtained with a second, validated method may also result from the inability to limit the vis-DRS light course to the target tissue. Ubbink et al.³⁰ tested the utility of vis-DRS to measure mucosal oxygen saturation of the bowel in pigs in vivo by comparing vis-DRS values to oxygen saturation values derived from simultaneously measured microvascular oxygen tension [see Sec. 3.5, Eq. (1)].^38,41 While oxygen tension increased with ${\mathrm{FiO}}_2$ and decreased promptly after euthanasia, vis-DRS values decreased with increasing ${\mathrm{FiO}}_2$ and remained relatively constant after euthanasia, suggesting that the probe measured Hb saturation in submucosal (venous) vessels. In breast tissue, for SDSs spanning 0.23 to 1.10 mm, Bydlon et al.⁴² reported a sensing depth of 0.5 to 2.2 mm. Liver has a much higher Hb content, i.e., absorption is about an order higher than that of breast tissue, but its scatter is about 1/3 to 1/2 of the scattering of breast tissue. For the SDS of 1.2 mm of our custom-made probe, we thus empirically estimated a penetration depth of 0.5 to 1.5 mm below the surface in our model. At this depth, liver tissue is relatively homogenous and without larger blood vessels.⁴² We thus expect that the values obtained with our setup truly reflected the oxygen saturation of the liver tissue.

4.3.

Effect of Liver Tissue Properties on vis-DRS Values

The absorption and scattering values obtained with vis-DRS in our study [Fig. 3(a)] were quite different from the average values found in the heart muscle, which contains large amounts of myoglobin (average ${\mu }_a\sim 8{\mathrm{cm}}^{-1}$ and ${{\mu }_s}^{\prime }\sim 12{\mathrm{cm}}^{-1}$, Gandjbakhche et al.,⁴³ Haggblad et al.³⁹) Liver tissue is highly absorptive due to bile that is produced by the organ and that is present intracellularly and in the small bile ducts. We established continuous biliary drainage to avoid biliary stasis and ensure that liver tissue properties remained constant throughout the experiments. Inclusion of biliverdin, the main pigment contained in bile, as an absorber into the Monte Carlo inversions, revealed that inversions that did not include biliverdin slightly underestimated the values for scattering at low levels of scattering and low levels of Hb. This suggests that during extreme hemodilution, i.e., at a low concentration of red blood cells, the tissue properties of the liver including its bile content contributed relatively more to light scattering than at higher levels where scattering was also determined by red blood cells. Gandjbakhche et al.⁴³ described for heart muscle that path length and spectral characteristics of the reflected light were greatly dependent on capillary blood content.

In an in vivo swine model, bile that was applied to the gut mucosa altered measured vis-DRS values depending on bile composition (Ubbink et al.³⁰). We suggest that, in our study, liver tissue properties (very high absorption coefficient but low scattering coefficient compared with other tissues) resulting from bile and high Hb content was responsible for the deviation of the measured vis-DRS values from the standard ${\mathrm{PO}}_2\text{-}{\mathrm{SO}}_2$ curve for pigs.³⁸ This deviation was systematic, and there was a good correlation of vis-DRS values with blood oxygen tension after correcting the vis-DRS values with a conversion formula. Corrected vis-DRS values clearly reflected the acute decrease in tissue oxygen saturation with reduction of blood flow to the liver in vivo (Fig. 6). More research is needed to determine whether the conversion formula has to be adapted for specific conditions like fibrosis or cholestasis.