2.1.

Spectral Camera Specifications

Four spectral cameras have been used in this study, which are based on two different acquisition techniques. The specifications of each camera and its sensor are presented in Table 1. One camera is a hyperspectral pushbroom camera (Tivita Tissue 8, Diaspective Vision GmbH, Germany), which holds a CMOS sensor and a dispersive element. Each captured two-dimensional image consists of a spatial axis $y$ and the wavelength information $\lambda$. During a scan process, 640 $y\text{-}\lambda$ images are acquired along the spatial $x$ axis, giving the third dimension of a complete data cube. The system is described in more detail by Mühle et al.¹⁴

Table 1

The sensor and camera optics specifications of the different spectral camera systems used. The last three parameters have been set during tissue analysis of the porcine organs. The specific bandwidths and spectral resolutions of the three snapshot mosaic cameras can be found in Table S1 in the Supplementary Material.

Three cameras are snapshot mosaic cameras using two different sensor setups. Two cameras (MQ022HG-IM-SM4X4-VIS and MQ022HG-IM-SM5X5-NIR, XIMEA GmbH, Germany) hold a $4\times 4$- and a $5\times 5$-mosaic pattern, which includes 16 and 25 wavelength bands, respectively. The $4\times 4$-mosaic sensor is sensitive in the visible (VIS) spectrum while the $5\times 5$-mosaic sensor is sensitive in the near-infrared (NIR) spectrum. The $5\times 5$-mosaic camera contains a longpass filter in front of the optics to cut off all wavelengths $\lambda <675\mathrm{nm}$ to avoid secondary bands. Thus each pixel in the mosaic pattern corresponds to one wavelength band. These two cameras are described in more detail by Wisotzky et al.^10,29 The third snapshot camera (CMS-C, SILIOS, France) holds a $3\times 3$-mosaic pattern with a slightly different setup. The center pixel of the mosaic pattern is sensitive in the complete sensor sensitivity range and behaves as intensity pixel, whereas the eight surrounding pixels (px) select specific wavelengths $\lambda$ in the VIS spectrum. All three snapshot cameras can handle 60 fps or higher.

In addition to the four camera systems, a spectrometer has been used as a reference system for the analysis of tissue and blood samples. The spectrometer (USB2000 + VIS-NIR-ES, Ocean Optics Inc., USA) with its beam path (QR400-7-VIS-BX, Ocean Optics Inc., USA) is an asymmetrical Czerny–Turner monochromator using a CCD detector. It has a spectral resolution of $1/3\mathrm{nm}$ and covers a wavelength range of 350 to 1000 nm.

2.2.

Imaging Setup

The acquisition setup with all four cameras and illumination spots is shown in Fig. 1. The illumination unit contains six 20-W quartz-tungsten-halogen spots (OSRAM Decostar 51 ALU, Osram GmbH, Germany) with an aluminum reflector and a defined beam angle and beam position (Fig. 1, No. 5). Each spot has a luminosity of 510 cd and a color temperature of 2800 K. All spots are mounted in a distance of 450 mm to the target and combined having an illuminance of 12.2 klux in that working distance (WD).

Fig. 1

The setup is shown during (a) calibration and (b) as sketch with the complete equipment: (1) pushbroom camera setup, 41-bands setup with (2) $4\times 4$-VIS camera and (3) $5\times 5$-NIR camera, (4) $3\times 3$-VIS camera setup, (5) six illumination spots, (6) spectrometer, (7) tissue sample, (8) heating unit, (9) thermometer, (10) cursor, and (11) perfusion machine for blood measurements.

All cameras (Fig. 1, Nos. 1 to 4) focus on the same target area, which is indicated by a cursor defining a circular region of interest (ROI) of $d=30\mathrm{mm}$ (Fig. 1, No. 10). Snapshot cameras acquire the entire wavelength information in one shot resulting in a spatial/spectral 2D representation of the dataset but with a reduced spatial resolution and fewer spectral bands (8 to 25 for the used cameras in this study). As the two Ximea snapshot cameras (Fig. 1, Nos. 2 and 3) are mounted on one static plate (stereo-like) and cover different spectral intervals, i.e., $\lambda =463$ to 638 nm for the $4\times 4$-VIS and $\lambda =693$ to 966 nm for the $5\times 5$-NIR camera, respectively, the data of both cameras are combined to one spectral dataset using pixel correspondences between the two cameras calculating a disparity map^30–33 (hereinafter referred to as multispectral 41-bands setup) and are then analyzed. The Silios snapshot camera (Fig. 1, No. 4) covers the same spectral interval as the $4\times 4$-VIS camera and furthermore covers the gap between the two Ximea cameras. Hence, it is used as a comparable $3\times 3$-VIS setup. This allows an estimation about the minimum required number of spectral bands (16 bands of $4\times 4$-VIS versus 8 bands of $3\times 3$-VIS) for robust organ surveillance. The pushbroom camera (Fig. 1, No. 1) captures the complete spectral dataset of a recorded slit in one shot. The second spatial component is created by moving the slit. The recorded data cube consists of 640 pixels in the traverse direction of the slit and 480 pixels in the direction of the slit. The angle between the illumination units and the different cameras is in the range of $\alpha \approx 25\mathrm{deg}$ to 35 deg. For reference data, a spectrometer (USB2000 + VIS-NIR-ES, Ocean Insight, USA) measures the object in the circular area (Fig. 1, No. 6). The spectrometer points toward a small region at the border of the circular area. As the spectrometer has no local resolution, the reference spectrum is averaged over the entire measurement area.

To obtain the reflectance spectrum from the measured raw data, each image has to be corrected according to

2.3.

Organ and Blood Analysis

We have analyzed tissues of typical donor organs (heart, lung, and kidney) to compare the four introduced spectral cameras. In addition, brain and blood have been analyzed as well. All organs used in this study have been considered healthy by the veterinarian present at the time of collection. In addition, one kidney affected with cysts is collected. All organs are retrieved from three female German Landrace pigs weighting 100 to 120 kg. The cyst-affected kidney is retrieved from the first pig, which is the only pig where both kidneys, a healthy and a cyst-affected kidney, have been collected. After death by exsanguination, all organs and the blood have been collected. Approximately 1.5 l of blood has been received from the carotid incision into a receptacle containing 10,000 units of heparin (Rotexmedica GmbH, Trittau, Germany). The blood has been stored at 4°C. All other organs have been removed en bloc and, with exception of the brain, flushed immediately with preservation solution (Custodiol, Dr. Franz Köhler Chemie GmbH, Bensheim, Germany). Afterward, the organs have been placed into a bag with preservation solution and stored on ice at 4°C. For blood collection, the time between exsanguination and storing on ice has been $\sim 2\min$. The process of organs and blood removal is described in more detail by Markgraf et al.²³ The warm ischemia time occuring between interruption of blood supply and storage on ice has been $\sim 44.5\pm 0.5\min$ for each porcine model. Before starting the measurement procedure, the organs have been removed from ice after $\sim 1$ up to 3.5 h cold ischemia time (CIT) and are heated up to the target temperature of $\sim 37°\mathrm{C}$. The heating process has been carried out for 25 min in a water bath. The procedure described in this paragraph has been adapted to simulate a state-of-the-art clinical transplantation procedure, where the spectral measurement would be performed 25 min after organ reperfusion.

2.4.

Measurement Procedure

At the beginning, reference measurements with a color chart (ColorChecker Classic, X-Rite Inc.) have been performed to set up the optimal camera and spectrometer settings, such as exposure time, focus, or frame rate, for each device (cf., Table 1). In addition, these measurements have been used to validate the spectrometer measurements as the colored tiles hold well-defined color spectra.³⁴ For each colored tile, the reflectance spectrum is known in the range of 390 to 1000 nm with step size of 10 nm.³⁵ Because of this validation, we can use the spectrometer measurements as ground truth or reference data for the organ measurements. The spectrometer measurements are performed by averaging 10 scans, a Boxcar width of 2, an integration time of 8 ms per scan, and a distance of 3.5 cm with angle of 12.5 deg to the target area having diameter of 1.95 cm. The software OceanView 1.6.3 (Lite, Ocean Optics Inc., USA) has been used to record and process the raw data.

Then all organs of each porcine model are measured successively in the order brain, kidney, lung, and heart. After each heating process, the organs are placed under the cursor (Fig. 1, No. 10) on a heating unit (Fig. 1, No. 8) to ensure a constant temperature during the measuring process. A thermal sensor (Fig. 1, No. 9) is placed inside the organs to control the core temperature during the measurements. For calibration purposes [cf., Eq. (1)], white and dark images are acquired before as well as after each organ measurement to achieve $I_{\text{white}}$ and $I_{\text{dark}}$, respectively. For acquisition of $I_{\text{white}}$, a white standard (99% reflectance Zenith, SphereOptics GmbH, Germany) is recorded simultaneously with all cameras and the spectrometer using the same illumination. Dark images are acquired after occluding the optics of all cameras and the spectrometer.

For each organ as well as the blood samples, three measurements have been performed with all cameras simultaneously. Each camera captures the ROI bordered by the cursor (Fig. 1, No. 10; cf., Fig. 4). One pushbroom measurement takes about 10 s. In parallel, 10 images have been acquired for each of the snapshot cameras. These 10 images are averaged to one dataset. Four different healthy organs of three pigs, one additional kidney affected with cysts as well as blood with three different oxygenation conditions have been scanned. Consequently, a total of 48 measurements has been performed for each camera: 36 healthy organ measurements, 3 cyst-affected kidney measurements, and 9 blood measurements. The blood conditioning experiments have been carried out in accordance to Markgraf et al.²³ and three different oxygenation saturation levels (98.6%, 92.8%, and 82.0%) have been set up. Since specular reflections are unavoidable in the ROI marked by the cursor, for the analysis, two smaller areas in the ROI are selected manually by biomedical experts considering the anatomical structures in each measured image of each camera, which contain no interfering artifacts. These two areas represent the similar location over all cameras per measurement. Each area has a minimum size of $21\mathrm{px}\times 21\mathrm{px}\times N$ bands, where $N$ is the number of possible wavelength bands dependent on the used camera. The measured sensor data of these ROIs are averaged for further evaluation in MATLAB R2019b (The MathWorks Inc., Natick, Massachusetts, USA).

Fig. 4

Spatially resolved HMSI images of the heart of the third pig model. (a) Multispectral $4\times 4$ VIS camera; (b) multispectral $5\times 5$ NIR camera; and (c) Multispectral $3\times 3$ VIS camera. One spectral band of the snapshot sensor data for (a) $\lambda =530\mathrm{nm}$, (b) $\lambda =746\mathrm{nm}$, and (c) $\lambda =515\mathrm{nm}$. (d) The intensity distribution of the TIVITA Tissue pushbroom camera image at $\lambda =750\mathrm{nm}$.

3.1.

Color Checker Measurements

Reference measurements with the color chart are applied to validate the spectrometer measurements using published reflectance data of the colored tiles. Several characteristic color tiles, showing different spectral distributions, have been scanned. Five representative reflectance data are shown in Fig. 2. The comparison between the spectra of the spectrometer and the known spectra of the color chart shows comparable results. The highest deviation between these spectra is 5%. In addition, two further color tiles are measured with the three camera setups to evaluate the comparability of the achieved reflectance data on a standardized, non-biological test object. The data recorded simultaneously by the four spectral cameras as well as the spectrometer have slightly different formats. Four spectral curves are achieved from the measurements (cf., Table 1): (1) spectrometer reference data ($\lambda =400$ to 1000 nm with step size $1/3\mathrm{nm}$), (2) hyperspectral pushbroom camera ($\lambda =500$ to 995 nm giving 100 bands in step size 5 nm), (3) multispectral 41-bands setup ($\lambda =463$ to 966 nm giving 41 bands), and (4) multispectral $3\times 3$-VIS snapshot camera ($\lambda =430$ to 700 nm giving eight bands). The camera setup results of the two color tiles are shown in Fig. 3. The pushbroom camera data as well as the 41-bands setup data follow the published color spectra as well as the ground truth spectrometer data. The multispectral $3\times 3$ camera data show more variability in the eight reconstructed data bands. In the following, these behaviors are presented and discussed in detail based on the organ measurements of the three analyzed pigs.

Fig. 2

Five different analyzed color tiles of the Color Checker Board. The continuous lines are measured with a spectometer in this study and the dashed lines are published reference measurements.^34,35

Fig. 3

Two color tiles, with defined colors (a) light skin and (b) magenta, analyzed with all four described setups. The two curves per imaging setup correspond to the two selected ROIs in each captured image.

3.2.

Organ Measurements

The measurements of the different porcine organs have been performed in a time range of approx. $t=2\mathrm{h}$, i.e., $t=2\mathrm{h}$ to $t=4\mathrm{h}$ after collection. Thus the CIT has been in the range of $t_{\mathrm{CIT}}=63\min$ and $t_{\mathrm{CIT}}=214\min$. The core temperatures $T_{\text{core}}$ of the healthy organs during the measurements are presented in Table 2. Organ core temperatures $T_{\text{core}}$ have ranged between $T_{\text{core}}=31.3°\mathrm{C}$ and $T_{\text{core}}=36.1°\mathrm{C}$.

Table 2

Organ core temperature Tcore (°C) recorded during the measurements of kidney, lung, heart, and brain in the three porcine models as well as blood temperature (°C) of different blood oxygen saturation levels (%). The different oxygenation levels of the blood sample are achieved via a perfusion oxygenator setup.23

A visualization of the captured data for the heart of the third animal is shown in Fig. 4. With the hyperspectral pushbroom camera, the largest field-of-view could be captured, followed by the multispectral $3\times 3$-VIS camera and finally by the multispectral $4\times 4$ VIS and $5\times 5$-NIR cameras. With all four camera systems, the measuring object field given by the cursor can be successfully recorded.

For each of the four porcine organs, a similar spectral behavior is captured through all four different acquisition methods in the corresponding spectral intervals of $\sim 450$ up to 700 or 1000 nm, respectively (see Fig. 5). All organs show a local peak at about $\lambda \approx 485\mathrm{nm}$, a local minimum at about $\lambda \approx 555\mathrm{nm}$, and a rising edge in the range of $\lambda \approx 600\mathrm{nm}$ to $\lambda \approx 630\mathrm{nm}$. The maximal reflectance intensity of the spectra is at $\lambda \approx 800\mathrm{nm}$. In addition, the different organs have specific spectral variations (see Fig. 5). The reflectance data averaged over all measurements results in an $\sim 50\%$ lower reflectance intensity for the heart over the entire spectrum compared to the other three spectra in an overall constant setting for all organs. Further, the reflectance variation over the entire analyzed spectrum is smaller for the porcine kidney and porcine heart samples compared to the lung and brain samples. Brain and lung data are comparable with specific differences. The brain spectra shows an intensity decrease in the infrared (IR) range at $\lambda =950\mathrm{nm}$, whereas it has slightly higher intensity in the visible range ($\lambda =460\mathrm{nm}$ to $\lambda =600\mathrm{nm}$) in comparison to lung data. The reflection spectra of the kidney are comparable to the brain and lung spectra and intersect them at $\lambda =600\mathrm{nm}$. It shows a flat trend in the NIR and IR range, with a decrease at $\lambda =950\mathrm{nm}$ similar to the heart spectrum.

Fig. 5

The spectral reflectance behaviors for all four analyzed organs averaged over all measurements presented of (a) spectrometer ground truth data, (b) hyperspectral pushbroom camera, (c) multispectral 41-band setup, and (d) multispectral $3\times 3$ camera.

In detail, differences through the spectral behavior of the organs exist over the complete analyzed spectra. While heart, kidney, and brain show a different decrease of intensity starting at $\lambda =950\mathrm{nm}$, the spectrum of the lung remains constant. Thus the lung shows a strongly increased reflectance from about $\lambda =950\mathrm{nm}$ compared to the other three organs. Furthermore, the rising band at $\lambda \approx 600$ to 630 nm is steep for heart, lung, and brain, whereas the increase in reflectance before and after that band is smooth. In addition, these organs show a differently pronounced local peak at $\lambda \approx 730\mathrm{nm}$ and a local minimum at $\lambda \approx 757\mathrm{nm}$. Considering the kidney dataset, the spectrum results are smoother, with a monotonic increase in reflectance from $\lambda \approx 560\mathrm{nm}$ up to $\lambda \approx 685\mathrm{nm}$. After this point, the reflectance remains relatively flat showing no pronounced local maxima or minima, different from the other organ samples.

All the obtained spectra are consistent with the spectral properties of common tissue chromophores in the range of 400 to 1000 nm. The primary light-absorbing compounds in tissue within this range are Hb and water. Depending on different oxygen conditions, a conformational change occurs in the Hb molecule, which leads to spectral changes in the UV–VIS–NIR range. The spectrometer data display in the range of $\lambda =380$ to 500 nm the extreme intense Soret bands.³⁶ The minima at $\lambda =555\mathrm{nm}$ and $\lambda =765\mathrm{nm}$ in the reflectance curve are characteristic bands of the porphyrin Hb molecule.³⁶ The flattening of the spectral curve from $\lambda =600\mathrm{nm}$ to $\lambda =740\mathrm{nm}$ as well as the formation of two minima in the range from $\lambda =500\mathrm{nm}$ to $\lambda =600\mathrm{nm}$ are spectral features of ${\mathrm{HbO}}_2$. For the lung data, the double peak of ${\mathrm{HbO}}_2$ is visible in the spectrometer, pushbroom, and 41-band data. The $3\times 3$-VIS data contain too few $\lambda$-bands to resolve the characteristic spectral properties. The observed local minima at $\lambda =760\mathrm{nm}$ and $\lambda =970\mathrm{nm}$ correspond to water-specific spectral properties, e.g., the absorption peaks of water.²⁵

Both the individual organ spectra of the camera measurements and the spectrometer measurements vary in relation to the reflectance intensity (i.e., intensity axis in Figs. 5Fig. 6Fig. 7Fig. 8–9). This behavior can also be detected in the spectrometer reference measurement of the three pigs. However, the measured and reconstructed 41-bands setup reflectance [cf., Fig. 5(c)] and the reflectance of the pushbroom camera [cf., Fig. 5(b)] lie within these reference ranges and correspond to the optical behavior of the organ. To quantify the similarity between the spectra of the spectral cameras and the spectra of the reference a normalized cross correlation (NCC), i.e., Pearson correlation coefficient, is applied (see Table 3). All NCC metrics are in the range of 0.97 up to 0.99 for the pushbroom camera and 41-bands setup data. The difference of about 1.5% for NCC values between the comparisons of spectrometer and pushbroom or 41-bands setup (0.995 versus 0.980) comes along the linear fitting between the single bands with wider band distances for the 41-bands setup data. In addition, the comparisons between the single camera setups are performed to characterize the comparability between the three different acquisition setups. The Pearson coefficients between pushbroom and 41-bands setup are in the range of $0.985\pm 0.005$, which shows a robust comparability between these two setups.

Fig. 6

For the porcine heart measurements, all curves acquired by the different spectral cameras—(a) hyperspectral pushbroom camera, (b) 41-bands setup, and (c) $3\times 3$-VIS camera—match the behavior of the reference spectrometer data (dashed lines). For each of the three camera acquisition methods, two ROIs are selected, reconstructed, and averaged over the three measurements. The resulting two ROIs are presented here.

Fig. 7

Porcine brain measurements of all three pigs both two selected ROIs are averaged to one spectral curve.

Fig. 8

Porcine lung measurements of the third pig at two different selected ROIs. These spectral curves are representative for the other two pigs as well.

Fig. 9

Porcine kidney measurements of (a) healthy kidney tissue from the second pig and (b) kidney with a superficial cyst from the first pig, both averaged over both measured ROIs. The spectral curves in (a) are representative for the other two pigs as well, which are omitted for readability. In (b), the spectrometer data of the healthy kidney of the first pig are plotted for comparability.

Table 3

NCC between the averaged spectral curves of the different acquisition methods. The curves of the pushbroom, the 41-bands, and the 3×3-VIS camera setup are linear fitted between the single bands. The number of data points indicates the used points for NCC calculation.

The measured and reconstructed reflectances of the $3\times 3$-VIS camera [cf., Fig. 5(d)] are different to the other two camera acquisition methods. The captured spectral data are located exclusively in the VIS wavelength range between $\lambda =425\mathrm{nm}$ to 700 nm. The first six wavelength bands in the range of $\lambda =425$ to 625 nm correspond to both the spectrometer reference data and the other two camera systems. Only for the heart measurements (cf., Fig. 6), the reflectance intensity of the first band is too high and do not fit the reference. Furthermore, the last two measured and reconstructed wavelength bands at approx. $\lambda =665\mathrm{nm}$ and $\lambda =700\mathrm{nm}$ show consistently too low reflectance intensities for all organs (cf., Figs. 5–9).

The NCC results of the $3\times 3$-VIS camera data substantiate these appearances (see Table 3). When comparing $3\times 3$-VIS camera with the reference spectrometer, the Pearson coefficient reaches $0.84\pm 0.01$ for kidney, lung, and brain measurements and 0.59 for heart measurements. Due to the data fitting for the pushbroom setup as well as for the $3\times 3$-VIS camera setup, the NCC is higher when comparing these setups (0.90 up to 0.97). Compared to the multispectral 41-bands setup, the NCC data are reduced, in a range of 0.74 to 0.83, as for both acquisition methods less data points are used for fitting.

The spectra of the porcine hearts are similar for all different recordings methods, i.e., spectrometer and different setups, and through all different measurements, shown in Table 3 and Fig. 6. The first data points of the spectra of the hyperspectral pushbroom camera ($\lambda \approx 500$ to 515 nm) as well as the first two spectral bands of the $4\times 4$-VIS camera of the 41-bands setup show a strong scattering or inaccurate behavior. This is visible for all measurements across all organs (cf., Figs. 7 and 8) for brain of all three pigs and lung of the third pig, respectively. Further, the 41-bands dataset consists of a large gap in the spectral range of $\lambda =640\mathrm{nm}$ and $\lambda =690\mathrm{nm}$. Therefore, in this region, the spectral behavior cannot be represented correctly, as this gap is the transition between the two combined multispectral cameras.

Especially, the multispectral snapshot cameras show some discontinuities in the reconstructed reflectance bands, as already described for the first two spectral bands of the $4\times 4$-VIS of the 41-bands setup. Further, the band at $\lambda =630\mathrm{nm}$ always underestimates the reflectance intensity. In contrast, all bands of the $5\times 5$-NIR represent the spectral organs behaviors with all local maxima and minima precisely.

In addition to the healthy porcine organs, one kidney affected with a cyst has been evaluated (see Fig. 9). This affected kidney is from the first pig and has been analyzed 3.75 h after organ collection with an organ core temperature of $T_{\text{core}}=37.0°\mathrm{C}$. Compared to the healthy kidney tissue of the first pig [cf., Fig. 9(a)], the spectral behavior of the cyst is completely different [cf., Fig. 9(b)], as a cyst contains a hydrous solution with cell components as well as hormones and electrolytes but almost no blood. This highly affects the spectral behavior. Between $\lambda =425$ to 900 nm, the spectrometer cyst data are in the range of $I_{\text{reflectance}}=0.35$. The local minimum at $\lambda =580\mathrm{nm}$ is visible in all camera setup data. Further, the local maxima at $\lambda =710\mathrm{nm}$ and $\lambda =815\mathrm{nm}$ as well as the local minimum at $\lambda =740\mathrm{nm}$ are formed by the pushbroom and the 41-bands setup. The NCC between the healthy kidney data and cyst data is 0.0385, which shows no relevant spectral correlation between these two types. Compared to the ground truth spectral cyst data of the spectrometer, the intensities of spectral data of the pushbroom camera as well as the 41-bands setup are slightly reduced. According to the NCC, the spectral curves of the pushbroom camera (0.996) and the 41-bands setup (0.930) are in good correlation with the reference spectra. Further, both setups are almost overlapping and comparable to each other with NCC of 0.926. The spectral intensity data of the $3\times 3$-VIS are slightly increased compared to the spectrometer data but follow the spectral trend as well.

3.3.

Blood Measurements

The blood has been included into the blood circuit after $t=6.75\mathrm{h}$ of storing on ice. The first blood measurement has started $t=7\mathrm{h}$ after storing. The temperatures of the blood sample during the measurements are presented in Table 2. The averaged blood temperature during all measurements has been $T_{\text{core}}=37.1°\mathrm{C}$.

The blood spectra show the expected behavior. There is a dependence on the degree of oxygenation of the blood with the reflectance characteristics. As shown in Fig. 10, the oxygenated blood curve has higher intensities in the near-IR ranging from $\lambda =600$ to 800 nm for the acquisition options of the hyperspectral pushbroom camera and the multispectral 41-bands setup. The multispectral $5\times 5$-NIR data of the 41-bands setup show higher variability along the spectral curve compared to the hyperspectral pushbroom blood data as well as compared to the spectral trends in the $5\times 5$-NIR organ measurements.

Fig. 10

Blood measurements of all three camera setups. For each setup, the two ROIs average over the three measurements are presented.

The individual behaviors for the different oxygenation levels with the characterized local minimum at $\lambda =760\mathrm{nm}$ for blood with low oxygen saturation is clearly visible for the pushbroom camera data. The 41-bands data follow the same trend, but it is less pronounced due to only 41 bands. The $3\times 3$-VIS camera data only cover the visual range and follow the expected blood spectra trend for the eight bands until $\lambda =675\mathrm{nm}$. The $\lambda =700\mathrm{nm}$ band is far too low in terms of reflectance intensity. For the sixth ($\lambda =615\mathrm{nm}$) and seventh band ($\lambda =665\mathrm{nm}$) of this camera, the differentiation between oxygenated and deoxygenated blood is visible. This has been also expected for the eighth band ($\lambda =700\mathrm{nm}$) but could not be proven from the data. No spectrometer data have been analyzed for the blood measurements as the spectrometer data have shown total reflections, which omit the spectral behavior of the different blood samples.