2.1.

Multispectral Camera for Fluorescence-Based Near-Infrared Angiography

To acquire real-time near-infrared fluorescence (NIRF) data for optical blood perfusion angiography, a custom multispectral camera (RGB-NIR camera) was used. The camera, outlined in Fig. 1, consists of an array of CCD photodetectors with a total resolution of $1600\times 1200\text{pixels}$ and a maximum frame rate of 40 fps. The CCD sensor has a dynamic range of 68 dB and a readout noise of $16{\mathrm{e}}^{-}$ with a maximum well capacity of $20{\mathrm{ke}}^{-}$. An array of pixel pitch-matched spectral filters is overlaid on top of the imager sensor for spectral (NIR and visible spectrum) sensitivity. The pixelated filter topology consists of a $2\times 2$ super pixel pattern repeated across the imaging array, where the individual pixelated filters are effectively interference filters optimized for sensing RGB and NIR ($>800\mathrm{nm}$) spectra. The spectral interference filters are achieved through a nanofabrication process, where multiple interleaved layers of materials with low and high dielectric constants are deposited on top of each other by physical vapor deposition. Because of an optimized nanofabrication process for realizing pixelated spectral filters, our filters have high transmission ratios ($>80\%$) and high optical densities ($>6\mathrm{OD}$), which are necessary for high fluorophore sensitivity. In addition to the pixelated layer, a notch filter (Semrock, NF03-785E-25), with an OD of $\sim 6$ at 780 nm, is added to the system to further suppress the fluorophore excitation light source from reaching the sensor. The compounded result of this architecture achieves a maximum ICG dye sensitivity of 10 nM under surgical light illumination (40 klux) with minimal spatial co-registration error among the spectral channels, i.e., color and NIR information. This leads to high blood flow detectability and co-registration of this information with the corresponding anatomic features. The camera is connected to a laptop via an ethernet cable to transmit 12-bits per pixel data at 40 fps. The laptop screen displays a color image, a fluorophore signal image in false color, and an overlaid version of both data frames through custom-developed software.

Fig. 1

Custom imaging setup used for optical blood perfusion angiography. The setup consists of the custom multispectral camera for FBA, a 780 nm laser and its optics, a visible-light LED panel, a laptop, and a heating pad for the animal to be placed. The laptop outputs a time vector for each of the six anatomic regions of interest. Points A and B in the inset denote the injection and first local maximum points, respectively.

2.2.

Imaging Setup

The imaging setup shown in Fig. 1 consists of the RGB-NIR camera, a laptop, a 780-nm laser (B&W TEK Inc., BWF2-780-0.8) and its optics, a visible-light LED panel (Genaray, LED-7100T), and a heating pad (Kaz, 7788-R). The laser is coupled to the custom laser optics through an optical fiber. The laser optics consist of a shaping line filter at 780 nm (Semrock, LL01-780-12.5), an aspheric condenser lens (Thorlabs, ACL25416U-B), and a diffuser (Edmund Optics, 47-994) to create a 10 cm uniform illumination pattern with $\sim 6\mathrm{mW}/{\mathrm{cm}}^2$ of excitation power. The LED panel produces 5 klux of visible light at the surface of the heating pad with a light temperature of 5000 K. Visible short-pass filters (3M, Cool Mirror Film 330) are placed on top of the LED panel to suppress any leakage of NIR light that could deteriorate the fluorescence signal acquired by the RGB-NIR camera. The RGB-NIR camera sits orthogonally above the heating pad at a distance of $\sim 60\mathrm{cm}$, with the laser optics and the LED panel placed on each side of the camera (Fig. 1). The anesthetized mouse is placed at the center of the heating pad in the supine position, as shown in Fig. 1, while keeping the heating pad at a temperature of $\sim 37°\mathrm{C}$ to ensure normal cardiovascular system behavior in the animal.

2.3.

In Vivo Small Animal Angiography

All animal studies were performed according to protocols approved by the Washington University School of Medicine Animal Studies Committee for humane care and use of laboratory animals. Optical blood perfusion angiography was performed on diabetic ($n=10$) and nondiabetic ($n=7$) mice. All mice underwent unilateral femoral artery ligation, and their arterial perfusion was assessed before and after ligation and through their recovery process (day 7 and day 14 postligation). The assessment was performed both with our FBA imaging system and with a laser-scanning Doppler imager for comparative reasons. The experimental approach comprised the three following steps.

2.3.2.

Laser-scanning Doppler blood perfusion imager of murine critical limb ischemia models

Diabetic and nondiabetic mice underwent serial arterial perfusion assessments of ischemic and nonischemic hind limbs at day 0 (preunilateral femoral artery ligation), day 1 (postligation), day 7, and day 14. Mice were anesthetized (using a ketamine/xylazine cocktail) and kept on an underbody warming pad set at $\sim 37°\mathrm{C}$, and hair was shaved from the distal abdominal and medial hind-limb regions. Each mouse underwent serial measurements using a high-resolution laser-scanning Doppler blood perfusion imager (Perimed, PeriScan PIM 3). Adductor, gastrocnemius, and hind-paw segments of ischemic and nonischemic hind limbs (Fig. 2) were analyzed by an independent investigator (who was blinded to murine cohort) using the manufacturer’s software (LDPIwin). Both mean and peak perfusion units for each anatomic region of interest were collected.

Fig. 2

Sample color image with outlines of the anatomic regions of interest. The image was taken on day 1, just after ligation. The left hind limb and right hind limb are labeled as ischemic (blue outline) and nonischemic (black outline), respectively. The hind limbs have been conceptually divided into three anatomic regions: gastrocnemius, adductor, and hind paw.

4.1.

Distal Arterial Perfusion During the Ischemic Recovery Process

Here, we summarize the differences between the gold-standard scanning laser Doppler technology and our FBA imaging system in evaluating hind-limb perfusion recovery in the setting of ischemia and diabetes. We present each technology’s ability to track the recovery process of each phenotype group, as well as data supporting our hypothesis that our FBA system reveals distinct differences in distal arterial perfusion during the recovery process of diabetic mice. The hind-limb perfusion ratios should be minimal immediately following unilateral femoral artery ligation since the ischemic vascular system has been seriously compromised. As time progresses, the ischemic hind limb should progressively recover arterial inflow and distal perfusion. This process is assumed to be impaired in the setting of diabetes.

To assess the recovery process, we have clustered the mice by diabetic status and experimental day. Table 1 shows the geometric centroids, equivalent to the cluster-member means, of each anatomic hind-limb perfusion ratio (adductor, gastrocnemius, and hind-paw) at day 0 before ligation, day 1 right after ligation, and days 7 and 14 after recovery for the RGB-NIR camera and the Doppler imager. Both technologies report maximum and minimum hind-limb perfusion ratios on day 0 and day 1, respectively, for all three anatomic regions. To more systematically evaluate the recovery process, we have computed the geometric distances between the centroids for days 1, 7, and 14 and the initial vascular-unaltered centroid for day 0. As expected, the geometric distance for both phenotype groups and for both technologies is greatest on day 1, when ischemic hind-limb perfusion is most compromised. Both technologies show that this geometric distance decreases with time in both phenotypes, confirming that both diabetic and nondiabetic mice are recovering from the induced ischemia. Interestingly, both technologies very similarly demonstrate that hind-limb perfusion on day 1 is more compromised in diabetic mice. These findings, combined with the results from the RGB-NIR camera on day 7 (described below), provide strong evidence that diabetic mice recover more slowly than their nondiabetic counterparts. Finally, the RGB-NIR camera detects a slower recovery, based on the geometric centroid distance, in both diabetic and nondiabetic mice on days 7 and 14 compared to the Doppler imager, possibly indicating that our FBA imaging system is additionally providing microvascular perfusion data that the Doppler imager is less sensitive to.

Table 1

RGB-NIR camera and Doppler imager centroid coordinates and recovery distances.

A more comprehensive view of these distributions is shown in Fig. 5, which offers a three-dimensional visualization of the perfusion ratio data for each phenotype for the RGB-NIR camera and Doppler imager. Figure 5 also shows the centroid locations for each experimental day cluster and their projections to the coordinate planes. These projections visually show how each of the three perfusion ratios agree with the findings in Table 1. Figure 5 also shows the two-dimensional profiles of gastrocnemius versus adductor and hind paw versus adductor. On both profiles, the cluster distribution trend can be appreciated for both phenotypes and for both technologies: day 0 data points cluster at the top right of the profile or coordinate plane, indicating high perfusion ratios; day 1 data points cluster at the bottom left, indicating low perfusion ratios; and day 7 and 14 data points gravitate toward day 0 with those on day 14 being the closest.

Fig. 5

Data plotted for RGB-NIR camera: (a) Doppler imager (b) for diabetic versus nondiabetic mice: 3-D visualization of the hind-limb perfusion ratios for each mouse with their respective cluster centroids and projections to the coordinate planes (left column) and 2-D profiles for gastrocnemius versus adductor (middle column) and hind paw versus adductor (right column).

4.2.

Dynamic Arterial Perfusion Variables Distinguish Diabetic from Nondiabetic Mice Prior to Ligation

We hypothesized that the subtle differences caused by the diabetic status on hind-limb perfusion should be sufficient to successfully distinguish diabetic mice from nondiabetic mice before induction of the CLI mode, and indeed, based on the dynamic arterial perfusion variables, we could identify the distinctive phenotypes. Based on the fact that diabetes causes microvascular complications, we hypothesized specifically that ICG propagation through the cardiovascular system would have different dynamics in different phenotypes as it approaches the hind-limb extremities. We support and frame this hypothesis with reference to the chemical properties of ICG and plasma: (a) ICG is known to bind to plasma and this binding is strongly dependent on the albumin protein content¹⁰ and (b) the albumin content in plasma is altered greatly in diabetic patients.¹¹

To evaluate our hypothesis, we assessed each mouse hind limb ($n=34$) separately, using the initial perfusion metrics on day 0 (three measures per hind-limb), before ischemia was introduced to the model. To frame these metrics in our hypothesis, the analysis scheme divides the hind-limb perfusion metrics by those of the hind paw, the most distal anatomic region. Figure 6 shows the clear cluster separation between the two phenotypes for the RGB-NIR camera, in contrast to the Doppler imager results, and the ability to draw a decision boundary between the two phenotype clusters computed with a support vector machine. The nondiabetic data points tend toward the top right section of the plot, while the diabetic ones tend in the opposite direction. This data separation does not appear with the data acquired from the Doppler imager.

Fig. 6

Phenotype clustering for the (a) RGB-NIR camera and (b) the Doppler imager. Unlike data from the Doppler imager, the RGB-NIR camera data allow a clear decision boundary to be drawn between the two phenotype clusters. Day 0 mice hind limbs, $n=20$ diabetic, 14 nondiabetic.

We attribute this difference, first, to the superior resolution and sensitivity of our FBA imaging system and its ability to capture dynamic perfusion variables in real time, and second, to the inherent architectural disparity coming from the different physical phenomena captured. The Doppler imager relies on measuring, for one point in space at a time, the backscattering of light caused by red blood cells. This measurement can suffer from heterogeneities, where the blood perfusion measurement can be affected by spatial variations and optical properties of the tissue.^12–14 This effect indicates that the perfusion signal can be compromised if the measurement site is changed, jeopardizing analysis of blood perfusion changes over time. The RGB-NIR camera, in contrast, measures the dye’s emitted fluorescence in an anatomic area of interest at a single point in time. This fluorescence is correlated not only to the blood volume and hence its perfusion but also to its chemistry, providing valuable knowledge about arterial perfusion angiography.