2.1.

Fluorescence Imaging Camera

The device consists of a camera head (25-cm long cylinder, maximum diameter 6.5 cm) and a medical computer unit connected together with a 3.5-m long cable. The weight of the camera head is approximately 1.1 kg. Inside the camera head (Fig. 2), light sources and a CCD camera are incorporated. Six high-power ($6\times 160\mathrm{mW}$ at 600 mA), narrow-band light-emitting diodes (LEDs, SMB735-1100 Epitex Inc., Japan) with a peak wavelength at 735 nm (spectral half-width 25 nm) are used for excitation of ICG. The fluence rate (irradiance) of the excitation light for a 30-cm working distance equals $2.3\mathrm{mW}/{\mathrm{cm}}^2$. Two additional LEDs (SMB830-1100, Epitex Inc.) emitting a wavelength of 830 nm (half-width 25 nm) are used to record the tissue anatomical images and to measure the tissue/lymph nodes’ absorption variation. The wavelength of 830 nm has been chosen to correspond to the maximum emission (between 825 to 845 nm in blood plasma) of ICG to ensure that the conditions for light propagation during diffuse reflectance measurement are very similar to those for the fluorescence photons. Fluorescence and absorption images are recorded in an alternating manner.

Fig. 2

Expanded view of the front part of the camera head (see text for more details).

The device is additionally equipped with three cold-white LEDs for on-demand illumination of the patient body, while other sources of visible light (lighting in the operating room) are switched off. An 8-bit noncooled CCD camera (Guppy F-038B NIR, Allied Vision Technologies) with enhanced NIR sensitivity (SONY ExView HAD CCD technology sensor ICX428ALL with maximum quantum efficiency of 61% at 600 nm and about 25% quantum efficiency at 820 nm, effective chip size $6.5\mathrm{mm}\times 4.8\mathrm{mm}$) was used as a detector array, equipped with a compact objective of 9-mm focal length (HF9HA-1B, Fujinon). For a working distance of 30 cm between the measured object and the front of the camera, the diagonal field of view equals 27 cm. The image resolution of the camera using the default mode (interlaced) is $768\times 492\text{pixels}$, allowing, however, for relatively slow image acquisition ($30/\mathrm{s}$) only. In our device, the camera runs in the so-called progressive readout mode (noninterlaced, $2\times 2$ binning) that enables it to increase the image acquisition frequency to $59\text{images}/\mathrm{s}$ with a $384\times 244\text{pixels}$ resolution.

The readout noise of the CCD camera at 20°C (without protection glass or standard filters) is given as $\sigma =64e^{-}$ and the pixel saturation capacity $C_{e.\mathrm{sat}}=25600e^{-}$. The corresponding dynamic range equals 400 (52 dB). From the linear relation between the variance and the mean of the photoinduced gray values (photon transfer method), we determined the overall system gain $1/K$ to be around $100e^{-}/\text{digital number}$ (DN). Since data transfer to the computer is restricted to 8 bit, whereas the analog-to-digital conversion is performed with a 12-bit resolution, the lowest 4 bits are skipped. Alternatively, a user-defined lookup table can be applied to convert the 12-bit data space in the camera head to the 8-bit data range for data transfer. The mean value of the signal-to-noise ratio at saturation (including spatial noise) is specified as 112 (41 dB). The camera has a preamplifier option for the analog video signal, offering a gain from 0 to 24 dB.

The optimal rejection of excitation light is important when designing a fluorescence imaging system. The broadband light source used here requires a filter to cut the longer wavelengths. In our camera, we placed high-quality ($\mathrm{OD}\ge 4$) short-wave pass filters (775 nm Techspec®, Edmund Optics) in front of each of the six 735-nm LEDs used for dye excitation. By combining this with the complementary long-wave pass filters on the camera side, we generated a custom bandpass filtering that minimizes excitation light leakage while maximizing recorded fluorescence light. In front of the objective lens, we placed two high-precision ($\mathrm{OD}\ge 4$, slope factor $<1\%$) interference filters (800 nm Techspec®, Edmund Optics). As the performance of interference filters strongly depends on the angle of incidence, we decided to additionally mount a colored, 3-mm thick glass filter (Schott RG-780) in front of the camera sensor. The effectiveness of the colored glass filters depends almost solely on the absorbing substrate thickness. Hence, photons with angles of incidence deviating from zero that might pass the interference filters will have a high chance to be absorbed by the colored filter. Hwang et al.³⁷ discussed in detail the excitation light leakage problem in in vivo fluorescence imaging, focusing on the importance of collimating optics for noise floor reduction when interference filters are used. Because of device cost and size limits, we decided not to implement expensive or big apparatuses like collimating lenses, image intensifier, or laser diodes. In our measurements, the excitation light leakage has played a negligible role. When illuminating tissue phantoms with our device, the average pixel intensity values (in the 8-bit intensity scale from 1 to 255 DN) in the absence of ICG did not exceed 2.5 DN, of which approximately 1.3 DN was the background measured in the absence of any light. The pixel intensity values of fluorescing lymph node phantoms were typically in the range between 100 and 200 DN. For all measurements presented throughout this paper, the CCD preamplifier gain was switched off.

The linear response of the camera system was confirmed ($r^2=0.990$) by illuminating a solid fluorescent block with a 735-nm light and recording fluorescence images in the functions of varying excitation light intensities corresponding to LED forward currents from 60 to 600 mA (in steps of 60 mA). The fluorescence signal (the mean value of all the pixels in the image) was plotted as a function of the diode current controlled via a digital-to-analog converter (NI USB 6009, National Instruments, USA).

The camera head is sealed with a cap made of polyphenylsulfonate, a material that is biocompatible, easy to clean, and suitable for autoclave sterilization. The central, transparent part of the front cap is made from an antireflective coated borosilicate glass window. In the operating room, the device is wrapped in sterile drapes, making it applicable and safe for intraoperative use. An accredited-notified body (Berlin Cert GmbH, Berlin, Germany) tested the device with respect to electrical safety (e.g., leakage currents) and other safety issues (according to DIN EN 60601 standard) relevant for its application during surgery.

A graphical user interface to control the device (Fig. 3) was developed under the LabVIEW (National Instruments, Austin, Texas) environment. The software allows for displaying four different images ($244\times 384$ pixels each) simultaneously: a raw fluorescence image (NIR emission), a reflection image representing tissue absorption (serving for the operator as anatomical grayscale image), a normalized fluorescence image (raw fluorescence image divided by absorption image), and a hybrid image. The term hybrid image describes the converted to visible “pseudocolors” normalized fluorescence image that has been overlaid onto the corresponding anatomical grayscale image. The user interface provides control of various system functionalities including: LED intensity adjustment, image display properties (contrast, brightness, and gamma correction) tuning, image archiving, and snapshot collection. Contrast and brightness (display) can be conveniently adjusted by the camera operator using the keyboard located on the camera head (cf. Fig. 1). The software allows for continuous recording of every fifth image acquired and the saving of the data as a tiff-stack. Adjusting contrast, brightness, or gamma correction (either using the user interface or keyboard located on the camera head) does not influence the recorded data, as only raw images can be saved to the file.

Fig. 3

User interface of the camera system. Upper left image: overlay of the pseudocolor fluorescent signal on top of the reflectance (830 nm) image; upper right: F/R ratio image; bottom left: raw fluorescence image (F); and bottom right: reflectance image (R).

2.2.

Lymph Node Phantom

To validate the feasibility of the device for real-time fluorescence imaging of lymph nodes, we constructed thin-walled ($300\mu \mathrm{m}$), 15-mm long hollow double cones (Fig. 4) made of polyoxomethylene. These lymph node mimicking cones were filled with a scattering fluid of the desired optical properties mixed with the fluorescent dye. The absorption coefficients and reduced scattering coefficients of the various liquid phantoms were derived from separate measurements of time-resolved transmittance on a 6-cm thick cuvette using a femtosecond Ti:Sapphire laser and time-correlated single-photon counting and applying the diffusion theory of photon transport for the homogeneous infinite slab with an extrapolated boundary condition.³⁸ The five cones were mounted in a holder ensuring their reliable and reproducible positioning. The holder was placed at the bottom of the glass reservoir (diameter: 18.4 cm, height: 9 cm) filled with the desired amount of a liquid tissue phantom consisting of water and whole ultra-heat treatment milk (3.5% fat). The milk to water ratio (21.6 to 78.4) was kept constant for all experiments, yielding a reduced scattering coefficient of ${\mu }_s^{\prime }=9.5{\mathrm{cm}}^{-1}$ at 780 nm. The schematic of the measurement setup is illustrated in Fig. 4.

Fig. 4

Experimental setup employed for measurements on lymph node phantoms. The scattering fluid was placed in a glass reservoir. Upper left corner: ultra thin-walled ($300\mu \mathrm{m}$) hollow double cone made of polyoxomethylene filled with scattering fluid and ICG to mimic lymph nodes.

The fluorescent dye we used is ICG-Pulsion (Pulsion Medical Systems SE, Germany), which is supplied as a sterile, water-soluble powder. If a water solution of ICG is administered intravenously, the dye molecules rapidly bind to the plasma proteins² which results in a red-shift of the absorption and emission peaks. In all measurements presented throughout this paper, the ICG dye was mixed with an excess of bovine serum albumin.

2.3.

Fluorescence Ratio Images

Assuming a typical ICG dose of about 1 mg injected within about 2 ml of solution results in a dye concentration of $500\mu \mathrm{M}$, which corresponds to the optimum concentration reported by van der Vorst et al.⁹ for SLN mapping in breast cancer. Although this concentration will be diluted in the tissue, we can expect high enough concentrations in the SLN that the reabsorption of the fluorescence light by ICG itself (concentration quenching) might play an important role (see Sec. 3). Only in the case of microdose investigations should the reabsorption of fluorescence light be negligible. Furthermore, the intensity of the ICG fluorescence signal can be substantially distorted by the interplay of absorption and scattering of the tissue. To retrieve attenuation-corrected fluorescence images with our device, the ratio technique (or F/R) was employed, i.e., the raw local fluorescence signal (F) of each pixel is divided by the corresponding local reflectance (R) at 830 nm. Background subtraction can be omitted here, since the observed fluorescence intensities are at least 50 times larger than the average background values. The F/R technique and its advantages have been reported for intraoperative fluorescence imaging by Ntziachristos et al.^39,40 Our device collects raw fluorescence images and reflectance images alternately with a repetition rate of 40 Hz (i.e., 25-ms acquisition time during illumination at 735 nm and 25 ms at 830 nm, respectively) and is capable of displaying the F/R images in real time. The nearly concurrent acquisition of raw fluorescence and reflectance images assures the same geometrical configuration of both measurements. In particular, the overall acquisition rate of 20 Hz is sufficiently short with respect to tissue movements related to the heart rate. Furthermore, the effects of ambient light fluctuations and natural vignetting cancel out in the ratio images.

2.4.

Analysis of Experimental Data

Since the illumination of the field of view is not homogeneous (up to 50% variation in the light intensity when illuminating with 735-nm LEDs), the acquired images require a correction procedure prior to analysis. For this purpose, a homogeneously fluorescent liquid phantom was illuminated separately by each set of LEDs (735 and 830 nm), where special attention was paid to assure identical geometry of the measurements on lymph node/tissue phantoms. Each pixel value in both raw fluorescence and reflectance images has then been adjusted resulting in an illumination variation below 5% for the 27-cm diagonal field of view in the corrected images.

In order to quantitatively determine differences between raw fluorescence images and F/R ratio images, the average photon counts in all double cones were calculated over an identical region of interest (ROI). The ROIs were chosen as squares of the size of 20 pixels, excluding the small square in the middle of each cone of 6 pixels’ size where the screws closing the cones pointing toward the camera head are present. In the vicinity of each cone, a ROI ($20\times 20\text{pixels}$) was chosen to determine the average local background signal. The contrast for each cone was calculated as the ratio of its average photon count signal and the corresponding background.

2.5.

Simulations

In order to theoretically investigate the influence of variations in the intrinsic (i.e., without dye) absorption of the lymph nodes on the fluorescence intensity, we used analytical solutions of the diffusion equations for fluorescence and excitation light. The geometry of superficial or excised lymph nodes was simulated by simply assuming a homogeneous semi-infinite turbid medium. For the case of a lymph node covered by a superficial layer of background tissue, the diffusion theory for a homogeneous semi-infinite medium containing a fluorescent and absorbing sphere was employed.⁴¹ In both models, illumination of the diffusive medium by wide-field excitation light was considered by summing up the simulation results for a two-dimensional grid of point-like light sources. Hereby, for each point source, the fluorescence and reflection signals were calculated for a $6\times 6{\mathrm{cm}}^2$ squared detector on a grid with step size of 2 mm.

3.1.

Concentration Quenching of Indocyanine Green

Figures 5(a)–5(c) show the images obtained by illuminating the five cones containing water solutions of ICG with different concentrations (1.0, 2.0, 4.0, 8.0, and $16\mu \mathrm{M}$). Here, only the bottom parts of the double cones were immersed in the scattering fluid. The camera was positioned above the lymph node phantoms in so-called 0-deg geometry. The distance between the front of the camera head and the top of the lymph nodes phantom was kept constant at 30 cm. The reflectance image [Fig. 5(a)] depicts a characteristic pattern of varying light absorption at 830 nm caused by the fluorophores inside the cones. Increasing the dye concentration 16 times (from 1 to $16\mu \mathrm{M}$) results in relatively minor signal intensity changes ($\sim 55\%$ decrease) in the reflectance image [Fig. 5(d)]. In the raw (uncorrected) fluorescence image [Fig. 5(b)], the cones filled with 4.0 and $8.0\mu \mathrm{M}$ of ICG solution appear nearly equally bright, brighter than the cone with the highest amount of fluorophore (the first one when looking from top). In the fluorescence image, which has been corrected for the absorption at the fluorophore emission wavelength [F/R ratio, Fig. 5(c)] the cone filled with the lowest amount of ICG appears the darkest, whereas the cone filled with the solution of ICG of the highest concentration appears the brightest.

Fig. 5

(a)–(c) Images (epi-illumination mode, 0-deg geometry) of five double cones filled with water solutions of ICG (the lowest concentration at the bottom and the highest on top). (a) Reflectance image, R. (b) Raw fluorescence, F. (c) Normalized fluorescence, i.e., F/R ratio. The bottom parts of the double cones are immersed in a diffusive fluid. (d) Normalized intensity for the five cones imaged.

The experimental data have been analyzed as described in Sec. 2.4. If one normalizes the intensities of the observed signals to that observed with the cone filled with $1.0\text{-}\mu \mathrm{M}$ ICG solution, then the normalized intensity of the cone filled with $2.0\text{-}\mu \mathrm{M}$ ICG is only 1.63 in the raw fluorescence image but is 1.86 in the ratio image. For the cone filled with $4.0\mu \mathrm{M}$ fluorophore, the corresponding values are 2.19 and 3.05, respectively. The lymph node phantom with $8.0\text{-}\mu \mathrm{M}$ dye appears in the uncorrected fluorescence image almost equally bright (2.23) as the one filled with $4.0\text{-}\mu \mathrm{M}$ solution, whereas in the ratio image it is much brighter (4.07). Finally, the cone carrying the highest concentration of the fluorophore ($16\mu \mathrm{M}$) is only 1.85 times brighter than the cone with the lowest dye concentration ($1.0\mu \mathrm{M}$) in the raw image, but the correction leads to a value of 4.15.

In the ratio images [Fig. 5(c)], the contrast between the cones and the surroundings as defined in Sec. 2.4 is significantly enhanced. In numbers, the enhancement with respect to the fluorescence image in Fig. 5(b) amounts to about 66%, 88%, 131%, 202%, and 270% for the five cones filled with 1.0, 2.0, 4.0, 8.0, and $16\mu \mathrm{M}$ of ICG, respectively.

3.2.

Simulating Tissue Heterogeneity

To simulate tissue optical heterogeneity, the same five cones described above were filled with a scattering liquid (milk and water, ${\mu }_a=0.03{\mathrm{cm}}^{-1}$, ${\mu }_s^{\prime }=9.5{\mathrm{cm}}^{-1}$) and different ink concentrations (10, 54, 113, 201, and 289 ppm), resulting in absorption coefficients of 0.1, 0.4, 0.8, 1.4, and $2.0{\mathrm{cm}}^{-1}$, respectively. The dye concentration in each cone was 200 nM. At this concentration, reabsorption of fluorescence by ICG is negligible. The cones were placed in the reservoir filled with a solution that mimics the background tissue optical properties (${\mu }_a=0.03{\mathrm{cm}}^{-1}$ and ${\mu }_s^{\prime }=9.5{\mathrm{cm}}^{-1}$). In the first experiment, only the lower halves of the double cones were immersed in the scattering solution to consider the case of superficial or excised lymph nodes (Fig. 6). Despite an equal dye concentration in each cone, the fluorescence images report different fluorescence intensities [see fluorescence image (F) in Fig. 6(b)] as a result of varying the intrinsic absorption of each cone. The varying absorptions can be seen from the diffusely reflected light at 830 nm [reflectance image (R) in Fig. 6(a)]. The reflection image was again used to correct the fluorescence image [F/R ratio in Fig. 6(c)].

Fig. 6

(a)–(c) Images (epi-illumination mode, 0-deg geometry) of five double cones filled with scattering fluid, ink, and ICG. The amount of ink was increased (from bottom to top), whereas the concentration of the fluorophore was identical (200 nM) in each cone. (a) Reflectance image, R. (b) Raw fluorescence, F. (c) Normalized fluorescence, F/R. The bottom parts of the double cones are immersed in a diffusive fluid. (d) Normalized intensity for the five cones imaged. (e) Simulated normalized intensities and F/R ratio assuming imaging of a semi-infinite homogeneous medium.

At each cone position, an ROI was defined as described in Sec. 2.4 to estimate the intensity as a function of the cone absorption. The results are shown in Fig. 6(d). In the case of the strongest absorption, the fluorescence intensity drops down to about 20%, whereas the reflection is much less affected. Accordingly, the fluorescence ratio can reduce the effect of intrinsic absorption in part only. In Fig. 6(e), the results of simulations are shown in which a homogeneous semi-infinite fluorescent medium with a plane surface was assumed for simplicity. The drop down of the fluorescence intensity with increasing intrinsic absorption is very similar to the experimental result, whereas the simulated reflection decreases more strongly than in the experiment. Hence, the simple model predicts a better correction by the fluorescence ratio than was observed experimentally, which will be discussed in the following section.

The absolute contrast was substantially higher in the ratio images than in the raw fluorescence images. For moderately absorbing cones (${\mu }_a=0.1$, 0.4, and $0.8{\mathrm{cm}}^{-1}$), the improvement was about 30%, for the cone of ${\mu }_a=1.4{\mathrm{cm}}^{-1}$, an improvement of 60% was seen, and for the cone with the highest absorption (${\mu }_a=2.0{\mathrm{cm}}^{-1}$), an improvement of 190% was observed.

In the next experiment, the cones filled with 200-nM ICG and various ink concentration were covered by 2 mm of scattering fluid of optical properties ${\mu }_a=0.03{\mathrm{cm}}^{-1}$ and ${\mu }_s^{\prime }=9.5{\mathrm{cm}}^{-1}$. All five cones are visible in both the fluorescence and the reflectance images (Fig. 7). However, the increasing intrinsic cone absorption reduces the diffusely reflected signal by a few percent only [Fig. 7(a)], indicating much less efficient absorption correction [Fig. 7(c)] than in the case of the uncovered cones. Figure 7(d) summarizes the relative changes of the peak intensities in the three images as a function of the cone intrinsic absorption.

Fig. 7

(a)–(c) Images (epi-illumination mode, 0-deg geometry) of five double cones filled with scattering fluid, ink, and ICG, covered by 2 mm of the scattering medium. The amount of ink inside the cones was increased (from bottom to top), whereas the concentration of the fluorophore was identical (200 nM) in each cone. (a) Reflectance image, R. (b) Raw fluorescence, F. (c) Normalized fluorescence, F/R. The bottom parts of the double cones are immersed in a diffusive fluid. (d) Normalized intensity for the five cones imaged. (e) Simulated normalized intensities and F/R ratio obtained by diffusion theory for a sphere embedded in a homogeneous semi-infinite medium.

Figure 7(e) shows the theoretical results obtained from diffusion theory for the semi-infinite medium with a spherical inclusion. The volume of the spheres was set to match the volume of the double cones used in the experiment. The decrease of the simulated signals with increasing absorption of the object corresponds very well to the experimental results. Experiments and simulations were also performed for larger thicknesses of the covering fluid on top of the cones. The thickness does not essentially change the relative decrease of the fluorescence intensity visible from Figs. 7(d) and 7(e). In contrast, the reflected intensity becomes almost completely independent of the cone intrinsic absorption when the top layer thickness is larger than 4 mm.

The absolute contrast was again better in the ratio images than in the fluorescence images. For the cones covered by 2 mm of tissue, the contrast was enhanced by 20%, 25%, 29%, 40%, and 47% for the cone absorption of ${\mu }_a=0.1$, 0.4, 0.8, 1.4, and $2.0{\mathrm{cm}}^{-1}$, respectively.