2.1.

Experimentation

Experimentation consisted of routinely imaging both tumor-bearing and control groups of animals using the LSCI technique over two weeks. All procedures were performed using protocols approved by the Johns Hopkins Animal Care and Use Committee.

2.1.2.

Setup and imaging protocol

Animals prepared with a thinned skull window were fixed in a stereotaxic frame. The cranial window was brought into focus using a lab jack with vertical manipulation. Magnification was kept constant at $1:1.3$ for all imaging sessions. Image acquisition was done using a 12-bit cooled charge coupled device (CCD) camera (Lumenera, Canada). Conventional reflectance images of the region of interest were obtained under white light illumination to serve as reference. Then, as shown in Fig. 1, a time stack of 80 raw laser speckle images was acquired at a constant exposure of 8 ms under red (632.8 nm, 0.5 mW) He-Ne gas laser (JDSU, Massachusetts) illumination. The final LSCI image was generated using the temporal speckle contrast processing scheme,³⁵ as shown in Fig. 1(b). The contrast value $K$ of every pixel was calculated as per

Eq. (1)

$$K(i,j)=\frac{{\sigma }_{80}(i,j)}{{\mu }_{80}(i,j)},$$

where ${\mu }_{80}$ and ${\sigma }_{80}$ are the mean and standard deviation, respectively, of intensities of every pixel $(i,j)$ across the acquired 80 images. The values were plotted to form a grayscale LSCI image, as shown in Fig. 1(c).

Fig. 1

Laser speckle contrast imaging of brain microvasculature. (a) Raw laser speckle images of an anesthetized rat’s brain are obtained through the thinned skull using a 12-bit cooled CCD camera under red laser illumination. (b) The acquired raw laser speckle images are processed using a temporal processing scheme to obtain an LSCI image. (c) The calculated LSCI image is a high-resolution wide-field image detailing microvessel structure.

2.2.

Data Analysis

The LSCI images obtained from longitudinal imaging experiments were processed for extraction of MVD. MVD was estimated as the number of nodes in a unit area of the LSCI image. A node was either an endpoint of a blood vessel or an intersection point (where a parent vessel branches into two daughter vessels). In essence, we transformed the problem of counting the number of vessel segments in a region to counting the number of intersection points and endpoints of vessels, since the latter count is unique to a defined area.

2.2.1.

Spatiotemporal study of MVD

We employed a custom written MATLAB (Mathworks, Massachusetts) program to count and calculate localized MVD values over the entire FOV. The program first displayed the obtained LSCI image, on which the user manually identified nodes, that is, intersection and end points of vessels. Figure 2(a) shows an example LSCI image with all nodes identified. The program proceeds to automatically calculate and represent the spatial distribution of MVD by counting the number of identified nodes within fixed or moving windows over the entire FOV. The manually identified nodes were counted and displayed using colored overlays. Counting of nodes was done within each square of a virtual overlaid grid, as shown in Fig. 2(b). We used a grid in which each square measured $110\text{pixels}\times 110\text{pixels}$ ($0.5\times 0.5{\mathrm{mm}}^2$). To obtain spatially continuous trends of MVD, as shown in Fig. 2(c), we also counted the number of nodes within a moving circular window around every pixel. If $(p_i,q_i)$ represents the complete set of $m$ identified nodes, the MVD at pixel location $(x,y)$ on the image was calculated as size ($N$), where

Eq. (2)

$$N=[(p_i,q_i)|{(p_i-x)}^2+{(q_i-y)}^2\le r^2]\forall i\in m.$$

Fig. 2

Estimation and display of MVD. (a) An example LSCI image of the rat’s cerebral vasculature, overlaid with a virtual grid for MVD estimation. Each square of the grid is $0.5\mathrm{mm}\times 0.5\mathrm{mm}$. The white dots indicate vessel intersection points and endpoints that have been manually picked toward MVD estimation. (b) The MVD count in pseudo color in each square of the grid. MVD is calculated by counting the number of vessel endpoints and intersection points in each square and multiplying by four (so that MVD is expressed per ${\mathrm{mm}}^2$). (c) A continuous spatial map of MVD values calculated in a circular $0.2\text{-}{\mathrm{mm}}^2$ neighborhood around every pixel. The black dots in (c) correspond to the white dots in (a).

For each pixel, we counted the number of nodes in a $0.2\text{-}{\mathrm{mm}}^2$ circular area (radius $r=55\text{pixels}$) around it to calculate the MVD at that pixel. Evaluating the MVD at every pixel in the image gave us a continuous MVD trend over the entire FOV. This information was color coded and overlaid on the LSCI images.

3.1.

LSCI Can Visualize Remodeling of Brain Microvasculature in Longitudinal Experiments

Figure 3(a) and 3(b) shows the LSCI images of rat brain microvasculature obtained on the day of tumor inoculation and 10 days after inoculation, respectively. They clearly demonstrate the ability of LSCI to visualize the changes in vascular architecture and the appearance of new vessels. The entire region of interest (ROI) has been divided into 15 squares with an area of $0.25{\mathrm{mm}}^2$ to enable a spatial comparison of vascular remodeling. The number of vessel segments was counted manually in each square. For example, in square (2,3), which lies above the inoculation site, the vessel count increased from 17 on day 0 to 29 on day 10, suggesting robust neovascularization. From Fig. 3, one can also observe that the original vasculature remains intact, thus allowing us to register the two images for comparison, but the addition of new vasculature makes it extremely difficult for any automated registration schemes to achieve the same.

Fig. 3

LSCI images reveal long-term vascular remodeling. LSCI images of the rat brain region of interest obtained (a) on the day of tumor inoculation and (b) 10 days after inoculation. The two images have been registered manually, and the entire area has been broken down into a $3\times 5$ square grid labeled (1,1) to (3,5).

Histology was used to confirm the presence and location of the tumor post mortem. Figure 4 shows H&E stained axial sections of two rats along with day 14 LSCI images to provide the readers an estimate of the size and location of the inoculated tumors with respect to the imaged ROI. Data from the tumor group was used for statistical analysis only if the tumor was visible and appropriately located under the imaging window. In an odd occurrence, one of the supposed tumor-bearing rats did not show a tumor in the histological sections, which implied that the inoculated cells had not developed into a tumor. Such data was discarded. Ten rats showed intracranial tumors and constituted the tumor group.

Fig. 4

Confirming tumor location under imaging window using histology. H&E staining was used to confirm tumor location. (a) and (d) show the LSCI images acquired of the region of interest before sacrifice. (b) and (c) clearly show the tumor on day 14 in the axial section with the box indicating the location of the imaging window with respect to the whole brain and tumor. The presence of tumor under the imaging window was used as a criterion for inclusion of data in the tumor group.

3.2.

Spatial Assessment of Microvessel Density over Tumor Growth Course

We demonstrated the capability of LSCI to quantitatively assess the microvascular structure and morphology before and after tumor angiogenesis in terms of MVD. MVD was estimated by manually picking intersection and endpoints of vessels in each LSCI image. Counting the nodes produced intuitive color-coded MVD maps, as in Fig. 2, which were overlaid on the original LSCI images. This allowed us to compare the extent of neovascularization quantitatively in terms of increase in MVD over the baseline. Figure 5 shows a comparison of one control (unperturbed) rat brain and one tumor-bearing rat brain on days 0, 7, and 14. These continuous MVD maps show a rapid increase in MVD in the tumor-bearing animal, while the control shows only minor variation. All MVD maps are scaled similarly. The MVD map of the tumor-bearing animal clearly showed a hot spot with a high degree of vascularity, seen as a red patch in Fig. 5. The peak MVD at this hot spot was $81{\mathrm{mm}}^{-2}$, in comparison to an average MVD of $19{\mathrm{mm}}^{-2}$ in the baseline image. The hot spot roughly correlated with the site of inoculation. However, since the tumor was fairly big on day 14, it remains to be seen how precisely the hot spot correlates with tumor location and size.

Fig. 5

Spatiotemporal comparison of MVD values in tumor-bearing and control rats. The images show continuous MVD maps overlaid on the cerebral vasculature of an animal from the unperturbed control group (left panel) and the tumor-bearing group (right panel) on days 0, 7, and 14 after baseline imaging and/or tumor injection. All images are scaled to the same color map (as indicated). The change in MVD over time in the control animal is small, whereas a steep increase in MVD is observed in the tumor-bearing animal.

3.3.

Temporal MVD Trends in Tumor-Bearing Rats and Controls

MVD values were estimated across the FOV in all 27 rats (10 tumor-bearing, six saline-receiving controls, and 11 unperturbed controls) over the five imaging sessions. Figure 6 shows the trend of mean MVD values over the entire ROI through the two-week course of tumor growth. The tumor group shows a sharp increase in MVD relative to day 0, while both the control groups witnessed an increasing-decreasing trend in MVD. The inset table reports the groupwise statistical significance ($p$ values) on each imaging day calculated using the Mann Whitney test. There is a statistically significant difference ($p<0.05$) between the MVD values of the tumor group and both the control groups on days 3 through 14. However, there is weaker evidence of the two control groups being different ($0.05<p<0.10$) on days 3 through 10. On day 14, there is strong evidence that both the saline-receiving group and the unperturbed group belong to the same distribution ($p=0.98$). On day 14, the tumor group exhibited relative MVD values of $1.24\pm 0.13$, which was significantly higher than the aggregated mean of both control groups of $0.88\pm 0.06$ ($p<0.02$). On day 14, absolute MVD values were also significantly higher in the tumor group ($24.40\pm 1.41{\mathrm{mm}}^{-2}$) than in the unperturbed control group ($17.64\pm 2.18{\mathrm{mm}}^{-2}$, $p<0.02$) or the saline-receiving control group ($13.33\pm 0.88{\mathrm{mm}}^{-2}$, $p<0.01$).

Fig. 6

Longitudinal variation of MVD over the course of tumor growth. Three groups—tumor ($n=10$), saline controls ($n=6$), and unperturbed controls ($n=11$)—were monitored using LSCI over two weeks since the day of inoculation (day 0). The plot shows means and standard error bars of the MVD averaged over the ROI and normalized to its corresponding baseline MVD value on day 0. The inset table shows results ($p$ value) of a Mann Whitney U test performed to compare the three experimental groups pairwise. The asterisk indicates statistical significance ($p<0.05$) between groups. The dagger indicates that the day 14 MVD of the saline and unperturbed groups belong to the same distribution.

4.1.

LSCI Can Visualize Remodeling of Brain Microvasculature in Longitudinal Experiments

LSCI has conventionally been used as a blood flow imaging technique, and hence LSCI users have used a spatial processing scheme that involves calculation of speckle contrast in the spatial domain; that is, $K$ is calculated in a sliding window (typically $7\times 7\text{pixels}$) across a single acquired raw laser speckle image. This scheme compromises on the spatial resolution and cannot discern the extent of new vessel formation. For assessment of microvascular changes that characterize tumor environments, a high-resolution temporal scheme needed to be utilized.^3,35 Figure 7(a)–7(d) shows a comparison of the difference in resolution as seen with the spatial and temporal processing schemes of the same raw LSCI data obtained on day 0 and day 3 in one of our experimental rats. We observed that the improvement in resolution enabled the identification of tumor-associated neovasculature, which comprises of microvessels less than 15 microns in diameter. We analyzed the enhancement provided by the temporal scheme in estimating MVD by comparing MVD values achievable using the temporal and spatial schemes in each square of a $7\times 10$ square grid overlaid on the ROI. This was done in 15 images (three rat brains over five imaging sessions), and the results are shown in Fig. 7(e). The median and quartiles over 15 images of the maximum density of nodes in each image were calculated and compared for the spatial and temporal schemes. The temporal scheme was shown to perform significantly better for imaging microvessels, and it was able to discern tumor-associated vascular remodeling. We have previously demonstrated the utility of LSCI for monitoring angiogenesis that accompanies wound healing in a mouse ear model,³⁷ and now, we demonstrate its utility to image vascular remodeling in the brain and skull. It must also be noted that the temporal scheme is sensitive to motion artifact, such as brain pulsations that could become more significant in bigger animals. Miao et al. have previously described a scheme of registering the raw frames to one another before speckle contrast is calculated, to address such motion artifact.³⁸

Fig. 7

Demonstration of the utility of LSCI with a temporal processing scheme to identify tumor-associated neovasculature. Baseline LSCI images of rat cerebral vasculature were taken using (a) traditional/spatial processing and (b) temporal processing. Equivalent (c) spatially and (d) temporally processed LSCI images were taken after the tumor was grown for several days. Note the improved resolution of (b) versus (a) and (d) versus (c). Also note the neovascularization evident in comparing (d) with (b). (e) A statistical comparison of the utility of the two schemes for microvessel identification. The median and quartiles are plotted for MVD values measured ($n=15$). Images have been scaled uniformly and linearly to improve print reproduction.

4.2.

Spatial Assessment of Microvessel Density over Tumor Growth Course

The LSCI modality, in conjunction with the MVD estimation and display techniques, enables serial assessment microvascular remodeling in the same rat over several days. This makes it possible to normalize the MVD values to the baseline MVD and study the temporal variation of MVD in each rat. We observed that there was significant variation in baseline MVD values of each rat of $20.47\pm 5.67{\mathrm{mm}}^{-2}$; that is, a standard deviation as high as 28% of the mean. This finding suggests that MVD trends need to be monitored independently for every rat, rather than being averaged over the entire cohort. Hence, LSCI-based monitoring provides an inherent advantage over classical terminal techniques of monitoring angiogenesis such as histology and immunohistochemistry, where it is impossible to monitor trends individually for every rat.

The large FOV of LSCI allows for monitoring of microvessels overlying the entire spatial spread of the tumor mass. As can be seen from Fig. 4, the size of the tumor on day 14 is comparable to the size of the imaging window. Spatial assessment of MVD showed regions of increased MVD or hot spots (see the red region in Fig. 5). While these hot spots appear in regions overlying the tumor, the precise correlation between the associated MVD values and the tumor location and size remains to be investigated.

At the same time, it must be appreciated that LSCI is a surface imaging modality and provides a 2D rendition of the ROI overlying the tumor. The depth of imaging is limited by the penetration of red (632-nm) laser in brain tissue. While use of a longer wavelength in the near infrared or infrared regime can improve the depth range, the same would compromise the in-plane spatial resolution, since the size of laser speckles is proportional to the wavelength.³⁹ Alternatively, LSCI could be complemented with other optical imaging technologies such as optical frequency domain imaging, which can resolve the tumor microenvironment depth-wise.²¹ In our experiments, the glioma cells were introduced as superficially as possible into the intracranial cavity, roughly 2 mm below the surface of the thinned skull, so that vascular changes extend to the surface as the tumor grows. For confirmation, MRI was done on two randomly chosen tumor-bearing rats on days 10 and 14. As shown in Fig. 8, the center of the tumor was confirmed from these images to be about 3 to 3.5 mm from the midline and about 2 to 2.5 mm from the surface of the thinned skull preparation. However, the tumor was also estimated to reach up to approximately 1.5 mm from the surface, as a hint of the tumor is seen in the topmost transverse section. The location of the tumor co-registered well with the thinned skull monitoring window.

Fig. 8

The site of tumor inoculation. (a) Coronal section stained with H&E. (b) Coronal T2 weighted MRI section. These images confirm that the glioma cells were inoculated at a depth of approximately 2 mm from the surface. (c) Transverse T2 weighted MRI sections confirm the tumor mass extends to within 1.5 mm from the surface (section D-D). All images shown were obtained on day 14.

Neovascularization is often measured by estimating vascular length or equivalently functional vascular density⁴⁰ (FVD). Spatial MVD maps were found to be significantly correlated (correlation coefficient $>0.98$, $p\ll 0.001$) to spatial FVD maps across all five imaging sessions in both tumor ($n=1$) and unperturbed control ($n=1$) rat brains. However, since our goal was assessing tumor environments, it was important to have a parameter such as MVD that was sensitive to short branching microvessels as are associated with neoplasia, rather than long nonbranching segments.

4.3.

Temporal MVD Trends in Tumor-Bearing Rats and Controls

When imaging the rodent glioma model through a thinned skull, the brain encountered three types of insults. First, the thinning of the skull triggered a wound healing response, which is accompanied by neovascularization. Second, the act of injecting a liquid mass (tumor or saline) into the brain using a Hamilton syringe created an insult that is expected to trigger an immune response of its own. Finally, our targeted insult consisted of introducing glioma cells into the brain, since the goal of the study was the monitor the effect of those cells. Our three groups of rodents—the unperturbed controls, the saline-receiving controls, and the tumor-receiving experimental group—helped us decouple the specific effects of each insult as pertains to vascular remodeling.

After a large amount of skull thinning on day 0, there was expectedly an immediate thermoregulatory and healing response that increased blood flow to the thinned skull within the FOV through redundancies, thereby causing the measured MVD values in the ROI to appear elevated in all three groups. On days 3 to 14, skulls were re-thinned only marginally. Hence, on these days, this marginal secondary insult did not produce as much of an increase in measured MVD as on day 0. As shown in Fig. 6, days 3 and 7 were marked with progressively increasing MVD, a result of long-term neovascularization that accompanies the healing skull. The difference in the relative MVDs among the three groups on days 3 and 7 are attributed to the difference in the injected mass. MVD values in the tumor group were significantly different from the controls, while the difference between the two control groups was only weakly significant ($0.05<p<0.10$), possibly due to a residual effect caused by the second type of insult. Day 10 witnessed decreasing neovascularization due to the first two types of insults, but the tumor group showed consistently high MVD values. Finally, on day 14, the two control groups showed the same level of vascularization ($p=0.98$), suggesting that the vascular remodeling effect of the first two types of insults had nearly completed, and the MVD values had returned to a true baseline, which was different from day 0 MVD values, since the latter represented the effect of significant bone thinning. On day 14, the MVD values in the tumor group were significantly higher than the values in both control groups, suggesting an exclusive effect of tumor growth and vascular recruitment. Figure 5 shows a representative spatial distribution of this confounded multi-insult vascular remodeling response over the two-week period. The intermittent increase in MVD in the unperturbed controls on day 7 can be attributed to wound healing, while the progressive increase in MVD in the tumor-bearing animal can be attributed to a combination of wound healing and tumor on day 7 but predominantly tumor on day 14.

Thus, as this paper demonstrates, skull thinning triggers a wound healing response that needs to be accounted for when conducting longitudinal studies. Other disadvantages of the thinned skull preparation include a limited ability to visualize only the surface of the brain and poor tissue depth penetration due to the surface imaging nature of LSCI. In addition, there is no objective way of ensuring the same skull thickness during different imaging sessions, which can result in variable image quality. However, a compelling reason to employ the thinned skull preparation is that it preserves the integrity of the brain microeonvironment, allowing the disease model to mimic naturally occurring physiological conditions. In the case of brain tumors, such conditions include progressively rising intracranial pressure as the tumor grows, as well as the chemical and immunological balance associated with an uncompromised brain.

4.4.

Potential Applications

The ability of LSCI of imaging tumor-associated vascular remodeling lends itself to various laboratory and clinical applications. This method could find application in anti-angiogenic drug studies, whereby the efficacy and spatiotemporal effects of the drug on vasculature could be studied. Through appropriate experimental design, it becomes possible to study not only the effect of anti-angiogenic drugs on tumors, but also their side effects on wound healing. Drug doses can be titrated based on the temporal rate of decay of MVD values in presence of the drug or the spatial outreach of vascular disruption over the wide FOV. In addition to high-resolution structural imaging, LSCI can image the blood flow through microvessels, thereby adding another dimension to angiogenesis research. Drug delivery to proximal and distal sites, especially in the context of new concepts like vascular normalization,^41,42 can be probed using the same platform.

In the rodent glioma model with an initial injection of 100,000 9L cells, the size of the tumor on day 14 is analogous to a clinical situation in humans, when surgical intervention is likely. Tumor resection is a delicate balance between removing as much tumor as possible and leaving behind as much healthy tissue as possible. In such a critical surgery, LSCI has the potential to assist the neurosurgeon with informative MVD maps of the FOV. Unlike complicated equipment like intraoperative MRI scanners,⁴³ the LSCI equipment is simple and can be easily integrated into the operating room. On day 14, absolute MVD values were also significantly higher in the tumor group than in the unperturbed control group ($p<0.02$) and the saline-receiving control group ($p<0.01$). This result, in conjunction with the presence of elevated MVD hot spots on the spatial MVD map, provides preliminary evidence that MVD could be used to discriminate tumors intraoperatively.

Further, it is possible to automate the estimation of MVD using advanced computational vessel segmentation approaches⁴⁴ to facilitate its objective use in the laboratory or clinic. Our study used manual identification of vessel intersection and endpoints as the gold standard to obviate any error associated with automatic approaches. White, George, and Choi have reported a method of segmenting vessels from LSCI images and subsequently skeletonizing the vessels to their respective single-pixel-thick centerlines to estimate vessel length.⁴⁵ Vessel intersection points and endpoints could be identified automatically as pixels on these vessel skeletons that have three neighbors and one neighbor, respectively, forming the basis for MVD calculation. However, the accuracy of such automated algorithms remains to be determined.