Optical oxygen saturation imaging in cellular ex vivo lung perfusion to assess lobular pulmonary function

: Ex vivo lung perfusion (EVLP) is an emerging tool to evaluate marginal lungs in lung transplantation. However, there is no objective metric to monitor lobular regional oxygenation during EVLP. In this study, we developed oxygen saturation (SaO 2 ) imaging to quantitatively assess the regional gas exchange potential of the lower lobes. Ten porcine lungs were randomly divided into control and donation after circulatory death (DCD) groups (n = 5, each). Lungs were perfused in cellular EVLP for 2 h, and multispectral images were continuously collected from the dorsal sides of the lower lobes. We examined whether lower lobe SaO 2 correlated with PaO 2 /FiO 2 (P/F) ratios in lower pulmonary veins (PV). The wet/dry ratio in lower lobes was measured and Monte Carlo simulations were performed to investigate the method’s feasibility. There was a significant correlation between lower lobe SaO 2 and the P/F ratio in lower PV (r = 0.855, P < 0.001). The DCD group was associated with lower SaO 2 and higher wet/dry ratio than the control group ( P < 0.001). The error of estimated SaO 2 was limited according to Monte Carlo simulations. The developed technology provides a noninvasive and regional evaluative tool of quantitative lobular function in EVLP.


Introduction
Ex vivo lung perfusion (EVLP) has been utilized to expand the donor lung pool in clinical lung transplantation using marginal criteria donors. In Lund-type EVLP, the perfusate containing red blood cells (RBCs) ensures high reliability of blood gas analysis for transplant suitability [1]. The PaO 2 /FiO 2 (P/F) ratio is a key criterion in the evaluation, with a threshold of 300 mmHg [2,3]. Although blood gas analysis from the left atrium (LA) is useful for the global assessment of double lungs, differential blood gas analysis from each pulmonary vein (PV), palpation, deflation test, and bronchoscopy are important tools in the assessment of regional and lobular lung function. Costa et al. reported that differential blood gas analysis provides a more accurate assessment of regional lung function than blood gas analysis in LA in clinical donor procurement [4]. Moreover, Freitag et al. developed a novel bronchoscopic method for gas exchange assessment of lobes and lung segments [5]. Both approaches, however, allow only intermittent evaluation, and the bronchoscopy technique is invasive and might interfere with the EVLP protocol, including deflation of the lungs. Currently, there is no continuous and noninvasive imaging technology to evaluate regional oxygenation in EVLP.
Hyperspectral and multispectral imaging (MSI) is a method providing spectrally resolved information at each pixel of an acquired image. MSI can be achieved by combining a camera and many different spectroscopic technologies such as multi-position filter-wheels, gratings and prisms, laser-scanning single point spectrographs, electronically adjustable tunable filters, and Fourier-transform imaging spectroscopy [6]. MSI was first developed for remote sensing in planetary science and astronomy [7]. In general, biological tissue is optically turbid media. The optical properties of biological tissue, which are light scattering and absorption coefficient, are affected by the pathological condition and biological activity. Therefore, MSI has been rapidly developed for medical applications, offering great potential for a technique of quantitative diagnosis [8]. Especially, the use of MSI in oxygen saturation (SaO 2 ) imaging has been focused on for the diagnosis of retina hypoxia and skin mottling [9,10]. Visible and near-infrared wavelengths, which range from 650 to 1000 nm, are easy-to-use, penetrate into tissue, and the light absorption sensitivity is altered by oxy-and deoxy-hemoglobin in blood flow. Moreover, for organ perfusion, Tetschke et al. demonstrated that MSI discriminates between perfused and non-perfused areas when monitoring SaO 2 during kidney perfusion [11]. The application of MSI to lung perfusion, however, has not yet been reported.
Basically, optical SaO 2 imaging depends on the difference in light absorption at two or more wavelengths [12,34]. The difference in light absorption is determined by the optical path length through the scattering in biological tissue. Therefore, the optical properties of the tissue are important for accurate quantitative diagnosis by MSI. Beak et al. investigated optical properties of non-perfused explanted piglet lung in different ventilation status at 632.8 nm [13]. Yi et al. measured the optical properties of deflated rat lung by inverse spectroscopic optical coherence tomography at 630 nm [14]. Jacques modeled the relationship between optical properties of various biological tissue including lung and wavelengths [15]. We hypothesized that SaO 2 imaging will reflect regional oxygenation of the lower lobes in cellular EVLP. Here, we developed a novel quantitative, real-time, and noninvasive method to image SaO 2 by MSI using visible and near-infrared light in a porcine lung EVLP model. The accuracy of SaO 2 imaging was analyzed using Monte Carlo simulation in the reported range of optical properties in lung.

Study design
Ten pigs were randomly divided into a control group and donation after circulatory death (DCD) group (n = 5, each). In the control group, lungs were perfused in EVLP following 1 h of cold ischemic time. In the DCD group, lungs were perfused after 90 min of warm ischemic time and a subsequent 5 h of cold ischemic time. During EVLP, MSI images of the dorsal sides in the lower lobes were continuously recorded. The following parameters were collected at 1 and 2 h: P/F ratio in LA, P/F in lower PV, and airway parameters. At 2 h, lung tissues were obtained to calculate wet/dry (W/D) ratios and lung weight gain.

Animal preparation
Males, crossbred between Landrace and Large White pigs (42-49 kg), were used. Tracheal intubation followed subcutaneous injection of 2 mg/kg xylazine and 20 mg/kg ketamine, and isoflurane of 1.5-3.0% was started. In the control group, lungs were procured in a standard fashion [16]. Two liters of Perfadex + nitroglycerin 10 mg/L was flushed antegrade in the pulmonary artery (PA), and 1 L of Perfadex was used for cold storage. In the DCD group, 500 units/kg heparin was intravenously administered, and after 3 min, pigs were sacrificed by intravenous injections of potassium chloride (2.0 mEq/kg). Following cardiac arrest, the pigs were kept in a supine position for 90 min, and lungs were procured in the same manner as in the control group. Blood was collected from a blood donor pig, and packed RBCs were prepared by a cell saver (XTRA; LivaNova Japan K. K., Tokyo, Japan). This study was approved by the Institutional Animal Care and Use Committees of the National Institute of Advanced Industrial Science and Technology (Approval number #2020-0336) and Tokyo Medical and Dental University (Approval number #A2020-100C). All methods were carried out in accordance with relevant guidelines and regulations.

Cellular EVLP
The organ chamber was made of optically transparent polycarbonate to capture MSI ( Fig. 1(A)). All optical instruments, including a light source, mirror, and hyperspectral camera (HSi-300; Gooch&Housego Inc., Ilminster, UK), were compacted under the chamber so as not to interfere with EVLP operation. Lungs were perfused according to the Lund protocol as previously reported [17][18][19]. Briefly, the system was primed with 2.0 L of STEEN solution, heparin 10,000 IU, imipenem 100 mg, and 500 mL of packed RBCs. Hematocrit (Ht) levels in the perfusate were maintained in a range of 10-15%. During the recondition phase, a gas mixture of O 2 , N 2 , and CO 2 was delivered to the membrane at flow rates of 2.7, 9.3, and 0.7 L/min, respectively. LA was left open. The flow rate increased in a stepwise fashion to a target flow of 70 mL/kg/min. Lungs were ventilated with the following settings when the lung temperature reached 36°C: Tidal volume (TV) 6 mL/kg, respiratory rate (RR) 7 breaths/min, positive end-expiratory pressure (PEEP) 5 cm H 2 O, and FiO 2 0.4. At 50 min, the following recruitment maneuver was performed: TV 10 mL/kg, RR 10 breaths/min, PEEP 5 cm H 2 O, and FiO 2 0.4 for 10 min. Then the O 2 gas supply to the membrane was stopped.

Blood gas analysis
For LA blood gas analysis, 2.5 mL of perfusate from each PV was collected in a small cup (10 mL) and mixed before being measured. For differential blood gas analysis of lower lobes, 2.5 mL of perfusate was obtained through a baby catheter located in each lower PV (outer diameter, 1.8 mm). PaO 2 and SaO 2 in the LA and lower PV at FiO 2 of 0.4, 1.0, and 0.21 were measured at 1 and 2 h by using a blood gas analyzer (ABL80 FLEX; Radiometer KK, Tokyo, Japan).

SaO 2 image processing from MSI
A xenon light source (MAX303; Asahi Spectra Co., Ltd., Tokyo, Japan) was used, and the light was guided by an optical fiber. A rod lens (RLQL80-2; Asahi Spectra Co., Ltd.) was attached to the outlet port of the fiber to uniformly illuminate the dorsal side of the lungs. The backscattering light was continuously captured throughout EVLP via a mirror by an acoustic-optic tunable filter (AOTF)-based hyperspectral camera with a capture time of 400 ms/image at the resolution of 0.3 mm/pixel (10 3 ×10 3 pixels) and the out-of-band rejection of -30 dB. A normal RGB image taken by a digital video camera (HDR-CX120; Sony Corp., Tokyo, Japan) is shown in Fig. 1(B). The MSI was created from 12 sub-images of each wavelength (Range: 590-913 nm, Fig. 1(C)). To create an SaO 2 image, two images at 650 and 800 nm were extracted from the MSI ( Fig. 1(D)). In this study, we employed the simplest method proposed by Polanyi et al. [12]. The relationship between SaO 2 and the backscattering intensity of the two lights from the tissue was formulated as follows: ∆R(x, y) = I 650nm (x, y)/I 800nm (x, y) = α × SaO 2 (x, y) + β where ∆R (x,y) is the proportion value of two wavelengths at coordinates of x and y of the image, and I 650nm (x,y) and I 800nm (x,y) are brightness values at wavelengths of 650 nm and 800 nm. et al. [12]. The relationship between SaO2 and the backscattering intensity of the two lights  The light absorption at 650 nm significantly depended on SaO 2 , while that at 800 nm was an isosbestic wavelength and did not depend on SaO 2 . The constant values of α and β depended on the light intensity, detection sensitivity, and optical properties of the lungs. The SaO 2 data of blood gas analysis from LA at 1 and 2 h were compared to ∆R in the LA outflow area where the perfusate flowed out ( Fig. 1(D)). There was a significant correlation between SaO 2 and ∆R (n = 60, r = 0.881, ∆R = 0.033 × SaO 2 -1.621, Fig. 2). Δ ( , ) = 650 ( , ) 800 ⁄ ( , ) = × 2 ( , ) + on the light intensity, detection sensitivity, and optical properties of the lungs. The SaO2 data 155 of blood gas analysis from LA at 1 and 2 h were compared to ΔR in the LA outflow area where 156 the perfusate flowed out (Fig. 1D). There was a significant correlation between SaO2 and ΔR

Correlation between lower lobe SaO 2 estimated by MSI and P/F ratio in lower PV
Typical MSI images and corresponding SaO 2 imaging at an FiO 2 of 1.0 are shown in Fig. 3(A). The SaO 2 was calculated according to the formula of the correlation between ∆R and SaO 2 (Fig. 2). The areas circled by solid lines indicate the lower lobes. The areas outside the solid lines reflect oxygenated perfusate coming from LA outflow. The histogram of lower lobe SaO 2 is shown in Fig. 3(B). In addition to a single peak at 100%, there was an isolated peak at 98% in the control group, whereas there were two peaks at 86% and 97% in the DCD group. Lower lobe SaO 2 calculated by MSI (SaO 2 MSI ) was defined as the median point in the histogram of each lower lobe. Then, lower lobe SaO 2 MSI were compared with corresponding P/F ratios in the lower PV.

Relationship between lower lobe SaO 2 MSI and W/D ratio
To collect lung tissue samples for W/D ratio assessment from the location corresponding to SaO 2 imaging, the lower lobe dorsal surface (white dotted line area, Fig. 4(A)) was equally divided into three areas: top, middle, and bottom. Then, each area was further divided into four equal subareas according to the gravity point of each area. Lung samples (Size; 1 × 1 × 1 cm, n = 6, each side) were obtained ( Fig. 4(A)). In parallel, SaO 2 data (FiO 2 = 1.0) at 2 h were collected from the area (Size: 1 × 2.5 cm) covering the corresponding two lung tissue sampling locations in each area ( Fig. 4(B)). The correlation between mean W/D value of two subareas and SaO 2 data in each area was analyzed. Lung tissue was frozen at -80°C and dried at 60°C for 24 h for W/D ratio calculation.

Monte Carlo simulations of light propagation in lungs
To verify the validity of SaO 2 imaging in lung tissue, Monte Carlo simulations were conducted. The simulation program proposed by Wang was used [20]. According to Jacques [15], the h for W/D ratio calculation.  reduced scattering coefficient of tissues was estimated by the following equation:

193
The simulation program proposed by Wang was used [20]. According to Jacques [15], the 194 reduced scattering coefficient of tissues was estimated by the following equation: where μ's = μs(1 -g); μs is the scattering coefficient (cm -1 ) and g is the anisotropy factor. In this 197 study, the g of lungs was 0.95 as a typical value [14,15]. In the equation, the wavelength λ is 198 normalized by a reference wavelength, 500 nm, to yield a dimensionless value, which is then  where µ s = µ s (1 -g) ; µ s is the scattering coefficient (cm −1 ) and g is the anisotropy factor. In this study, the g of lungs was 0.95 as a typical value [14,15]. In the equation, the wavelength λ is normalized by a reference wavelength, 500 nm, to yield a dimensionless value, which is then raised to the power b, called the 'scattering power'. This term characterizes the wavelength dependence of µ ′ s which is the reduced scattering coefficient (cm −1 ). The factor a is the value µ ′ s (λ = 500 nm), which scales the wavelength-dependent term. However, the µ ′ s of the lung varies in the literature. The ranges of a and b were 8.1-25.4 and 0.53-0.98, respectively [14,15,21]. Then, µ ′ s (λ = 650 nm) and µ ′ s (λ = 800 nm) varied from 6 to 22 cm −1 and 5 to 20 cm −1 , respectively. Therefore, in this study, we performed Monte Carlo simulations by using the entire range of reported µ ′ s values of the lungs to investigate the possible errors in SaO 2 estimation depending on the optical properties of the lung tissue. Here, we considered that the reduced scattering coefficient of the lung, µ ′ s Lung can be described by a synthetic reduced scattering coefficient of perfusate (µ ′ s Perfusate ) and skeleton (µ ′ s Skeleton ). Skeleton is defined as a component of lung parenchyma without pulmonary intravascular perfusate. The formula is as follows: where k (0 ≤ k ≤ 1) is the mixing ratio of perfusate and skeleton. Then the scattering and absorption coefficients of perfusate were calculated by the following equations [22,25]: where Ht is the hematocrit (0 ≤ Ht ≤ 1), V is the mean corpuscular volume, and σ s is the scattering cross-section of RBC. In this study, a value of 50 µm 3 was used for V as reported in porcine RBC [22,23]. The value of 44 µm 2 for µ ′ s (λ = 800 nm) is derived from Mie theory [24], where   the relative refractive index between RBC and extracellular fluid was 1.04 and the anisotropy factor of RBC was g = 0.99 [22]. Equation (6) was proposed by Prahl [25], and e(λ) is the molecular extinction coefficient for the wavelength λ in cm −1 /mol/l. mHb is the molecular weight of hemoglobin (64,500 g/mol). tHb is the total hemoglobin of perfusate in g/dL. In this study, the mean ± standard deviation of Ht and tHb were 0.127 ± 0.09 and 3.9 ± 0.9 g/dL, respectively. Then, in the simulation, the two pairs of (Ht = 0.1, tHb = 3 g/dL) and (Ht = 0.13, tHb = 4 g/dL) were used in Eq. (5) and Eq. (6), respectively. Furthermore, e(λ) is dependent on SaO 2 as follows: where, e HbO 2 (λ) and e Hb (λ) are the molecular extinction coefficients of oxy-and deoxyhemoglobin, respectively. In Monte Carlo simulations, the following values were used: e HbO 2 (650 nm) = 368 cm −1 /mol/l, e Hb (650 nm) = 3750 cm −1 /mol/l, e HbO 2 (800 nm) = 816 cm −1 /mol/l, e Hb (800 nm) = 762 cm −1 /mol/l. The SaO 2 Perfusate (0 ≤ SaO 2 Perfusate ≤ 1) is the oxygen saturation in the lung, where SaO 2 Perfusate is considered as the SaO 2 of the mixed blood of PA and PV. In the simulation, three values of 0.8, 0.9, and 1.0 for SaO 2 Perfusate were used. The simulation model is shown in Fig. 5(A). The number of photons was 10 6 . µ ′ s Lung in Eq. (3) is shown in Fig. 5(B). Since µ ′ s Perfusate was larger than µ ′ s Skeleton , µ ′ s Lung decreased as k decreased. In addition, µ ′ s Lung fluctuated because µ ′ s Skeleton varied from 2 to 30 cm −1 at 650 nm and from 1 to 27 cm −1 at 800 nm. The value becomes higher as the tHb increases.
∆R in the simulation was calculated by dividing the diffuse reflectance intensity at 650 nm by the one at 800 nm. The SaO 2 in the lung was finally calculated by the following equation: where ∆R k=1 means the ∆R in perfusate simulated at k = 1.

Statistical analysis
Statistical analysis was performed using JMP version 15.2.1 (SAS Institute Inc., Cary, NC, USA). Data are expressed as a median and a range (minimum-maximum). Correlations between two sets of parametric data was analyzed by Pearson's correlation coefficient. Student's t test was used for parametric data. The breakpoint of the P/F ratio vs. lower lobe SaO 2 was identified by piecewise linear function fitting. A value of P < 0.05 was considered statistically significant.

Trend of SaO 2 imaging
The typical trend of SaO 2 images at 1 h is shown in Fig. 6. In the recondition phase, where the PO 2 of PA was 229 (188-348) mmHg, most of the area in both groups was red, corresponding to a SaO 2 of 100%. Following the shutoff of O 2 gas supply to the membrane, the color in the control group was consistently red, except the green part in the top area at an FiO 2 of 0.21. In contrast, the color in the DCD group changed depending on the three FiO 2 . The top area tended to be blue, while the middle and bottom areas were either green or red.

Significant correlation between lower lobe SaO 2 MSI and P/F ratio in lower PV
There was a significant correlation between lower lobe SaO 2 MSI and P/F ratios in the lower PV (FiO 2 = 0.4, 1.0, or 0.21, n = 120, r = 0.591, P < 0.001, Fig. 7(A)). Next, subgroup analysis was performed for each FiO 2 . After having identified a breakpoint, a linear correlation was analyzed by using a value below the plateau. The maximum linear correlation was observed for an FiO 2 of 1.0 ( Fig. 7(B)-7(D)).

Significant differences in SaO 2 MSI and W/D ratios of three areas in the lower lobes
The top, middle, and bottom SaO 2 MSI in the DCD group were significantly lower than those in the control group (Fig. 8(A), P < 0.001). The top, middle, and bottom W/D ratios in the DCD group were significantly higher than those in the control group (Fig. 8(B), P < 0.001). In addition, the top SaO 2 MSI and top W/D ratio were compared with those in middle and bottom SaO 2 MSI . ∆SaO 2 MSI of the top area, defined as top SaO 2 MSI -middle SaO 2 MSI , was significantly lower than the ∆SaO 2 MSI of the bottom area in both groups (control, P = 0.038; DCD, P = 0.043, Fig. 9(A)). In contrast, there was no difference between the top ∆W/D ratio and bottom ∆W/D ratio in the control group. The top ∆W/D ratio was significantly lower than the bottom ∆W/D ratio in the DCD group (P = 0.003, Fig. 9(B)).

SaO 2 error analysis
Simulated SaO 2 in the lung are shown in Fig. 10(A). The error from the true value (SaO 2 Perfusate = SaO 2 Lung at k = 1) became larger as the value of k became smaller. SaO 2 Lung was underestimated   as the µ ′ s Skeleton became larger. The error disappeared as SaO 2 Perfusate became smaller. The mean error ∆SaO 2 , defined as mean SaO 2 Lung -SaO 2 Perfusate , is shown in Fig. 10(B). The mean error ∆SaO 2 was logarithmically reduced as k increased.

319
The mean error ΔSaO2 was logarithmically reduced as k increased.   smaller. The mean error ΔSaO2, defined as mean SaO2 Lung -SaO2 Perfusate, is shown in Fig. 10B.

319
The mean error ΔSaO2 was logarithmically reduced as k increased.

343
In the relationship between lower lobe SaO2 MSI and P/F ratio at FiO2 of 1.0 (Fig 7B),

344
the P/F ratio reached a plateau at 500 mmHg, which corresponded to an SaO2 MSI of 95%. This 345 finding is different from that using a pulse oximeter for arterial oxygenation: there is a plateau SaO2 MSI derives from a mixture of both venous and arterial blood. For example, brain SaO2 348 using an optical tissue SaO2 sensor, ForeSight Elite (Edwards Lifesciences Corp., CA, USA),

349
shows a plateau of SaO2 at 78% [28,29]. The mean ratio of venous/arterial mixture (m) for the 350 brain has been determined to be 0.7 [28]. When SaO2 in venous blood is 69% [29], the plateau of the PA/PV mixture (m) in lungs is 0.5 [30]. Pure SaO2 of PA was measured by using the

Discussion
The most significant finding in this study was that reduced lower lobe SaO 2 MSI significantly correlated with low P/F ratios in lower PV and high W/D ratios of the lower lobes. This suggests that the proposed optical technique of SaO 2 imaging is a suitable for measuring extravascular lung water during cellular EVLP. Another important finding was that the correlation between lower lobe SaO 2 MSI and P/F ratio in the lower PV was the strongest at FiO 2 of 1.0 compared with FiO 2 of 0.4 and 0.21. This finding is consistent with our previous studies [17,26]. The P/F ratio at FiO 2 of 1.0 had the greatest area under the curve to detect non-suitable lungs. These findings are explained by the difference in the shunt fraction at each FiO 2 .
In clinical EVLP, double lungs with a questionable single-side lesion are frequently perfused, and the EVLP team has to decide whether both lungs, a single lung, or neither are suitable for transplant. Our group demonstrated that the ultrasound monitoring of lung surface (CLUE) is useful for isolated lobe evaluation during clinical acellular EVLP [27]. However, SaO 2 imaging might be better to continuously monitor all of the lung surface area than ultrasound, because CLUE intermittently collects images from a limited number of locations. Therefore, the finding of an isolated single-side blue color spot or multiple and asymmetrical blue color spots in SaO 2 imaging could result in better detection of a single-side non-functioning lung. As there is no established method to diagnose pulmonary embolism during clinical EVLP, it may be a better fit for this technology.
In the relationship between lower lobe SaO 2 MSI and P/F ratio at FiO 2 of 1.0 ( Fig. 7(B)), the P/F ratio reached a plateau at 500 mmHg, which corresponded to an SaO 2 MSI of 95%. This finding is different from that using a pulse oximeter for arterial oxygenation: there is a plateau of SaO 2 at 100%, because a pulse oximeter measures SaO 2 of arterial blood only. In contrast, SaO 2 MSI derives from a mixture of both venous and arterial blood. For example, brain SaO 2 using an optical tissue SaO 2 sensor, ForeSight Elite (Edwards Lifesciences Corp., CA, USA), shows a plateau of SaO 2 at 78% [28,29]. The mean ratio of venous/arterial mixture (m) for the brain has been determined to be 0.7 [28]. When SaO 2 in venous blood is 69% [29], the plateau point of brain tissue SaO 2 is calculated as follows: m × SaO 2 in venous blood + (1-m) × SaO 2 in arterial blood = 0.7 × 69% + 0.3 × 100% = 78%. Similarly, MSI in our study captured backscattered light coming from both PA and PV. Hall et al. reported that the estimated ratio of the PA/PV mixture (m) in lungs is 0.5 [30]. Pure SaO 2 of PA was measured by using the perfusate sample from the outlet of the membrane oxygenator in our cases, and was 83%. When healthy lungs (the control group) are considered, the SaO 2 of pure PV will be 100%. Then, the plateau value of SaO 2 MSI is estimated as 0.5 × 83% + 0.5 × 100% = 92%. The difference in the SaO 2 MSI plateau value between this estimation (92%) and our measured data (95%) might be explained by the potential variation in m of the lung superficial portion, where the MSI signal was collected, compared with whole lungs, including both superficial and deep portions. Another possibility is that SaO 2 MSI may be underestimated based on the results of the Monte Carlo simulations as discussed below. An additional important finding was that there was a significant difference in lower lobe SaO 2 MSI among the three areas: the top SaO 2 MSI in the lower lobes was significantly lower than those of the other areas. In addition, there was no difference in W/D ratios among the three areas in the control group, and anatomically, central and large-size PA was distributed in the top area, which might be because m is lower in the top area. These findings should be considered when interpreting SaO 2 imaging, especially when this technology is clinically applied. In this study, the dorsal side of the lower lobes was selectively evaluated by MSI technology. In early research by West et al., the 'gravitational model' explains that the dorsal side of normal lungs has higher pulmonary blood pressure than the ventral side [31]. In addition, Niikawa et al. reported that there is a heterogeneous manifestation of ischemia reperfusion injury (IRI) during EVLP, with variations between upper and lower lobes [19]. In a supine position, the lower lobes have worse pulmonary function than the upper lobes. Based on these findings, a new setting of MSI monitoring the lower lobe selectively was developed in our study. When considering the significantly larger size of the lower lobe compared to upper lobe in pig lungs, selective SaO 2 monitoring of the lower lobes might be justified to detect signs of IRI.
In this study, the SaO 2 MSI was estimated by using the correlation between ∆R and the SaO 2 in perfusate samples from LA (Fig. 2). Ideally, the SaO 2 of perfusate inside the lungs should be compared with MSI images. However, in reality, it is impossible to collect perfusate samples from individual locations of lungs. Therefore, the degree of this discrepancy in SaO 2 was investigated by Monte Carlo simulations. The simulation results indicated that the error between SaO 2 MSI and true SaO 2 values in the lungs was limited as k (the volume ratio of perfusate/perfusate + skeleton) increased. In this study, as the first report of SaO 2 imaging of lung, we demonstrated the feasibility of the most basic two-wavelength SaO 2 imaging. The combination of two-wavelength band-pass filter and a digital camera can do the same imaging without using a hyperspectral camera. As a future prospective, three-wavelength oximetry using two isobestic wavebands [32,34], the theory of multispectral oximetry [33,34], and deep spectral learning approach [35] might provide more accurate SaO 2 MSI in EVLP. According to the gravitational model, ventilation/perfusion (V/Q) becomes less than 1 in the dependent area, suggesting that perfusate relatively increases with decreased alveolar size [31]. Therefore, it is expected that the k value in the dorsal side is high, suggesting that the data of SaO 2 MSI in this study included minimum discrepancy. This finding is consistent with the small difference in the SaO 2 MSI plateau value between the estimated SaO 2 (92%) and our measured data (95%).
This study has several limitations. First, porcine lungs were utilized with a limited case number. Further studies using human lungs are warranted. Second, the depth of backscattering light penetration was limited. According to a report by Stolik et al. [36], the estimated penetration depth in human lungs is in a range of 1-10 mm. Therefore, SaO 2 MSI data might reflect oxygenation in a relatively superficial portion of the lungs. However, based on the significant correlation between SaO 2 MSI and both P/F ratios in lower PV and W/D ratios, SaO 2 MSI might be a good surrogate for lower lobe pulmonary function. Third, the correspondence between SaO 2 image data and lung tissue for the W/D ratio can include a small gap.

Conclusion
A novel, noninvasive optical SaO 2 imaging system for cellular EVLP was developed using a porcine lung model. Lower lobe SaO 2 imaging significantly correlated with the P/F ratio in lower PV. This technology provides noninvasive, continuous, and regional evaluation of lobular function in EVLP.
Funding. National Institute of Advanced Industrial Science and Technology (AIST EDGE Runners).