About 30% of all cancers among women were breast cancer in the United States in 2017 [1]. Monitoring the prognostic clinicopathologic factors (patient’s age, tumor size, histological grade, axillary lymph node involvement, the presence of vascular invasion, estrogen-progesterone receptor status, and human epidermal growth factor overexpression) [2] is critical to enhance life quality and to reduce medical expenditure. Several groups using immunohistochemistry have reported that the correlation of tumor size with microvessel density is controversial with a positive [2], negative [3], or no correlation [4]. In particular, Ullrich et al. monitored microvessel density and tumor vessel size over a few days to demonstrate the potential of an MRI imaging-based approach to validate anti-angiogenic treatment effects [5].

To resolve the controversy mentioned above and to ultimately help design a suitable tumor treatment strategy, several medical imaging approaches have been studied by exploring tumor angiogenesis in both clinical and preclinical models. These approaches include Doppler ultrasound [6], micro-magnetic resonance imaging (μMR) [7], micro-computed tomography (μCT) [8], photoacoustic tomography [9], fluorescence microscope [10]. However, ultrasound, μMR, and μCT cannot show a single vessel due to their limited resolution [11]. Fluorescence microscopy has been widely used to study angiogenesis in tumor models at the resolution of a single vessel; however, it requires systemic labeling of the vasculature through intravenous injections, which leads to limitations in daily longitudinal studies [12, 13].

Compared to the techniques above, optical coherence tomography (OCT) based angiography (OCTA) can provide label-free, high spatial-temporal and depth-resolved blood vessel morphological information. A potential application of OCTA is preclinical intravital cancer imaging [11]. Vakoc et al. demonstrated a compelling approach to study tumor biology in a mouse model using OCT which included tumor margin drawing in three dimensions as well as a viability image map and a combined map with lymphan-giography [14]. Moreover, they showed the longitudinal changes in several vessel parameters (tumor volume, intra-tumor vessel length, fractal dimension, tortuosity, and mean vessel diameter) in response to cellular targeted (diphtheria) and vascular targeted (DC101) treatment.

However, those vessel parameters were monitored only during tumor regression after chemotherapy but not during tumor growth. Therefore, this study aims to investigate the correlation between tumor volume and tumor vascular morphometrics including vessel skeleton density (VSD) [15], vessel diameter index (VDI) [15] and vessel complexity shown by fractal dimension (FD) and lacunarity [16, 17] not only during tumor regression due to chemotherapy but also during its growth from the beginning of tumor cell inoculation by using OCTA.

A spectral domain OCT system based on a Mach-Zehnder interferometer was developed to perform the OCTA. The light source (SLD, DenseLight Semiconductors Incorporated, Singapore) provides a center wavelength of 1310 nm and a FWHM of 140 nm. A linear-in-wavenumber spectrometer (Bayspec, USA) was integrated with an InGaAs line scan camera (Goodrich Corporation, USA) which has a 12 bit-depth, 2048 pixels and a maximum line rate of 76 kHz [18]. The lateral resolution was measured to over 21 μm with an OCT phantom (APL-OP01, Arden Photonics, UK). The axial resolution was measured to 13 μm in the air. The sensitivity was measured to 78.8 dB at a depth of 200 μm. The exposure power at the sample site was measured to be 10.66 mW which does not cause skin damage and is within the American National Standards Institute (ANSI) safety limit [19]. XY-axis galvanometers were synchronized with a camera using a retriggerable data acquisition board (NI PCIe-6353, USA) and a frame grabber (NI PCIe-1429, USA). The OCT code for the data acquisition and signal processing was written with Visual C++ and Microsoft Foundation Class (MFC). A CPU based parallel computing library (OpenMP in Microsoft visual studio 2008), a vector computing library (Intel integrated performance primitives, IPP 8.0), and multi-threading structure using Microsoft visual studio 2008 enabled the display of the OCT image in real-time [20].

To enhance the blood flow decorrelation signal, the A-line rate was empirically set to 30 kHz. The scanning protocol was determined to have a field of view (FOV) of 5 mm × 5 mm with 256 lines in each B-scan (fast axis); 1000 locations in the C-scan (slow axis) and B-scans were repeated four times at the same C-scan location. Therefore, the cross-sectional OCT image speed was ~117 fps, and thus, a volume dataset with four repeated B-scans was acquired for approximately 34 seconds at one time. The C-scan sampling density of 4.2 was more than the Nyquist criterion of 2, while the fast transverse scanning (B-scan) sampling density was ~1.07, and thus, the decorrelation value was calculated along with the slow transverse scanning (C-scan) to reinforce the blood flow contrast.

A dorsal skinfold window chamber (DWC) model was applied to balb/c/nu (female; ~8 weeks old; body weight, 22 g) mice to make an ectopic breast cancer model. The DWC surgery was performed according to the protocol reported by Palmer [21]. Three animals were divided into three cases (no tumor, treated tumor, and untreated tumor) to find a correlation between tumor blood vessel morphometrics and tumor volume. All the animals were taken care of in individual cages with food and water ad libitum during the experiment.

Around 230,000 13762-MAT-B-III (CRL-1666, ATCC, Manassas, VA) rat breast cancer cells in 20 μL of McCoy’s 5A medium (ATCC, Manassa, VA) were inoculated between the fascial layer and the dermis for the treated tumor and untreated tumor mice while the no tumor mouse received the same amount of McCoy’s 5A medium without tumor cells. The tumor size was measured by its horizontal (W) and vertical (L) lengths from the epithelial layer within the window chamber with a Vernier caliper and then was converted to the tumor volume according to the following equation: [22]. The procedures in this study were approved by the Institutional Animal Care and Use Committee of the Gwangju Institute of Science and Technology.

As shown in Fig. 1, the OCTA recording was started from the seventh day after the DWC surgery to allow for a full recovery from the surgery. The OCTA image was taken on a customized imaging mount to reduce the motion artifacts. In details, motion artifact could be minimized through the following procedure. The imaging mount consists of thick acryl plates to be used for animal bed and a hard aluminum plate to fix dorsal chamber part. Nude mouse lies on its side on the acryl plate. The nasal mask is fixed by using an acryl cover case. Three extra nuts and a precision hex nut driver are utilized for tightly fixing the dorsal chamber and aluminum plate together. The animal was anesthetized before imaging with 1.5% isoflurane mixed air gas (200 sccm flow rate). The heart rate and arterial oxygen saturation were recorded by a mouse pulse oximeter (MOUSEOX, STARR, USA). EtCO₂ and FiO₂ were measured by a gas monitor (B40, GE Healthcare, UK). The body temperature was maintained with a heating pad (SRFG-203/10-P, OMEGA, USA) attached to the imaging mount. For the treated tumor case, cyclophosphamide (40 mg/kg body weight) was given to the animal intraperitoneally every other day seven times starting from Day 6 after the tumor cell inoculation while the untreated tumor case received the same amount of distilled water. Therapeutic initiation was determined by monitoring the tumor growth of around 20 mm³. Every OCTA was acquired at 10 minutes after the initial anesthesia to stabilize the physiological state. The OCTA was taken every day until fifteen days for the no tumor case, eighteen days for the treated tumor case, and eleven days for the untreated tumor case. By using a smartphone camera, a daily digital image was also captured right after taking the OCTA map.

Inverse fast Fourier transform was applied to the real-valued spectral interferogram obtained from the spectrometer (for which the B-scans were repeated four times at the same C-scan location [15]) so that the three dimensional OCT complex valued data were extracted. By using the complex-valued data, we adopted a complex differential variance (CDV) [23] in MATLAB (R2013b) to acquire the vascular morphology map. To enhance the decorrelation signal, the CDV decorrelation value was calculated along with the inter-frame direction instead of the inter A-line direction [24]. To efficiently separate the blood flow decorrelation signal from the background signal, the cross-sectional map of the structural OCT was weighted to the decorrelation map as an adaptive thresholding value [23]. A step down exponential filter along with the depth was used to reduce the shadow artifact on the cross-sectional decorrelation map [14]. Although OCTA can provide depth-resolved angiography in detail, we used the daily OCTA maximum intensity projection (MIP) map to estimate the vascular morphological parameters with ease.

Before quantifying the vascular morphology, we defined the region of interest (ROI) by considering a characteristic vessel pattern on the whole daily MIP maps for each case (no tumor, treated tumor, and untreated tumor) as shown in Fig. 2(a). Within the user-defined ROIs on the daily MIP maps, we extracted the vessel area map and vessel skeleton map for each case. An intensity threshold and a pixel size threshold for all the cases were applied to get the vessel area map. However, vessel skeleton maps are likely to distort the evaluation of the vessel skeleton density near the large vessels compared to the capillary vessels as shown in Fig. 2(d) [15]. Therefore, we carefully selected the ROI along with a characteristic vessel pattern by looking over the whole daily vessel area maps for each case to avoid these undesirable skeleton patterns.

First, the vessel skeleton density (VSD) is computed as the ratio of the pixels registered as the vessel skeleton to all the pixels within the user-defined ROI on the vessel skeleton map as follows: [15]

where S(i, j) and R(i, j) refer to the white pixels and the total pixels within the ROI in the vessel skeleton map (Fig. 2(c)), respectively. (i, j) are the pixel coordinates in the image map with a size of an n × m pixel array. Because VSD is not dependent on the change in vessel diameter, it is said to be more sensitive to the perfusion changes at the small vessel than at the large vessel compared with the vessel area density (VAD) [15]. Therefore, we analyzed the VSD of the OCTA instead of analyzing the VAD due to the lateral resolution limitation of our home built OCT system.

By using both the vessel area map and the vessel skeleton map, the vessel diameter index (VDI) is calculated to estimate the averaged blood vessel caliber as follows: [15]

where A(i, j) and S(i, j) refer to the white pixels within the ROI in the vessel area map and vessel skeleton map, respectively (Figs. 2(b) and 2(c)).

To obtain the fractal dimension (FD) and lacunarity values, a box counting method in the FracLac plug-in (Karperien, A. version 2.5) [16, 17] was used on the vessel skeleton map within the ROI with a black background scan. FD describes the degree of complexity a pattern has by evaluating the degree of space filling in it [16, 17, 25] Fractal dimension, D_g, is calculated from N_𝜖 = A𝜖^-D_g which can be changed to

where N_𝜖 denotes the number of boxes which include the white pixels of the vessel skeleton map with respect to each scaled box size 𝜖 in each grid g laid at each different location. Simply, FD is 1 for a line and 2 for a rectangle in two dimensional geometry. Therefore, for a complex pattern like blood vessel morphology, FD is normally estimated as between 1 and 2 in the two dimensional space. Mean fractal dimension (FD) is the average of D_g along with the grid laid at each different location as follows: [17]

where N_g denotes the total number of grids involved.

Lacunarity describes the vessel heterogeneity by quantifying the variance of an image pattern [16, 17]. Here, the term “invariance” reflects how well an image pattern is maintained for similarity even though the image is rotated by 90 degrees or is translated. Lacunarity λ_𝜖,g is calculated as follows:

where CV_𝜖,g denotes the coefficient of variation with respect to each scaled box size 𝜖 in a series of grids g. μ_𝜖,g and 𝜎_𝜖,g refer to the mean and standard deviation obtained from the number of pixels with respect to the scaled box size 𝜖 in each grid g, respectively. Finally, lacunarity λ is the average value obtained from all the λ_𝜖,g collected.

Figure 3 shows a few of the daily OCT angiograms and their corresponding color photos for the no tumor, treated tumor, and untreated tumor cases. The FOV of the OCTA is indicated by a white dotted square with a size of 5 mm × 5 mm. The primary criterion for selecting these ROIs in each case was to consider its characteristic vessel pattern with the naked eye throughout a total of the daily OCT angiograms. Although they have different sizes due to the daily change in tumor size, therefore, it was easy to compare the variation of the vascular morphology within each landmark boundary in each case during the observation period. In the no tumor case, two ROIs were selected by visually inspecting the characteristic vessel shape as indicated by a red dotted line. For no tumor case, especially, both VSD and VDI were calculated as the summation of two values obtained from the two ROIs, while both the FD and lacunarity were calculated by averaging two values obtained from them. The full daily OCT angiograms for each case are described in further detail in the appendix.

We first compared the VSD with tumor size throughout the growth and regression of the tumor (Fig. 4). Overall, the VSD was found to have a positive correlation with tumor volume. In other words, it increases as the tumor grows and decreases as the tumor regresses its volume due to chemotherapy. In the treated tumor case, the maximum value of the VSD appeared on Day 7, and it started to decrease on Day 8, which is one day earlier than the time when the tumor started to regress its volume (Fig. 4(a)). For the untreated tumor case, the VSD showed a plateau after Day 6, which might reflect vaso-destruction followed by tumor necrosis (Fig. 4(b)). The corresponding daily OCT angiograms support the change in VSD for the untreated tumor case (Figs. 3(c) and A3). In the no tumor case, the VSD gradually decreased (Fig. 4(a)), but the change is relatively small compared to those in both the treated and untreated tumor cases.

The change in blood vessel diameter also showed a positive correlation with the change in tumor volume but to a much lesser extent (Fig. 5). For example, the VDI in the treated tumor case increased steeply from Day 0 to Day 8, while the value went down to approximately ten on Day 10 and then oscillated between 9.5 and 11 from Day 10 to Day 18 (Fig. 5(a)). Meanwhile, the VDI in the untreated tumor increased suddenly on Day 6 and maintained its level for another two days. Then, it slightly dropped on Day 10 possibly due to the loss of the blood vessels along with the horizontal direction (Figs. 3(c) and A3). For the no tumor case, the values of the VDI fluctuated between 7 and 10 (Fig. 5(a)).

Figure 6 shows the comparison of the vessel complexity parameters (i.e., FD and lacunarity) with the tumor volume throughout the growth and regression of the tumor. Traditionally, the FD has been known as a promising indicator to estimate tumor vessel complexity or abnormal vessel pattern [13, 25]. The change in FD has an explicitly positive correlation with the change in tumor volume. In the treated tumor case, the FD value increased until Day 7, and later, it went down (Figs. 6(a) and A2). The untreated tumor also showed an increase in FD until Day 6, and then, it reached a plateau. This is in agreement with the results using the intravital microscope [13, 25]. In the no tumor case, the FD value decreased gradually due to the disappearance of the blood vessel, which could result from the blood flow obstruction caused by the weight of the installed dorsal chamber during the experiment.

Lacunarity is the counterpart of the FD, and highly dense or severely sparse blood vessels are prone to have a lower value of lacunarity which corresponds to a homogeneous pattern in the tumor vascular network. Even if it did not show a strong correlation with the tumor volume, the lacunarity of the treated tumor case seems to correlate fairly negatively with the tumor volume. For example, the lacunarity decreased as the tumor grew because of a more homogeneous pattern, while it went back to its initial value during the tumor response to chemotherapy due to a less homogeneous pattern (Figs. 6(c) and A2). For the untreated tumor case, new blood vessel growth (which means tumor angiogenesis) from Day 1 to Day 4 increased the lacunarity, but then, the lacunarity decreased on Day 6 due to the reduced variation of the vascular pattern (Figs. 6(d) and A3). Lacunarity continued to increase from Day 6 to Day 11 due to the occurrence of vaso-destruction resulting in a more complex vasculature like a heterogeneous pattern (Fig. A3). For the no tumor case, lacunarity on Day 7 abruptly went up due to the sparse blood vessels, and later, it steeply went down on Day 14 because the blood vessels became denser than before (Fig. A1).

We demonstrated that the VSD, VDI, and FD have a strong positive correlation with tumor size throughout tumor growth and treatment while lacunarity has a negative correlation. In addition, despite the resolution limitation of our home built OCT system (~21 and ~13 μm in lateral and axial directions, respectively), we found that both the VSD and FD can be useful to investigate the tumor vascular network compared with the VDI (Figs. 4-6).

Vakoc et al. showed that the intratumoral blood vessel length (which was monitored longitudinally by three-dimensional OCTA) decreased remarkably or increased slowly in response to either VEGFR-2 blockade or diphtheria toxin administration, respectively [14]. Similarly, we showed that the change in the VSD observed by the OCTA MIP maps fairly well followed the change in tumor size in response to cyclophosphamide administration. Especially, we found that the reduction of the VSD occurred one day before the tumor regression in the treated tumor case (Fig. 4(a)).

In addition, we found that there is an asymmetric change in the VSD during the time between tumor growth and its regression for the treated case. The value of the VSD on Day 6 was higher than that on Day 13 (13 vs. 4 [×10^-3 μm]) even though the tumor volumes on both days were similar (~20 mm³). It means that the tumor vasculature on Day 6 was denser than that on Day 13, which can be explained as follows. The VSD increases rapidly during tumor growth because the angiogenesis occurs before the tumor starts to grow while the destruction of the tumor vasculature occurs earlier than the regression of the tumor volume during chemotherapy. As a result, the VSD has potential as an indicator for the early prediction of tumor response to treatment like the vascular reactivity in our previous report [26].

Meanwhile, the measurement of the blood vessel diameter may help interpret abnormalities in the tumor vasculature [10] because changes in the diameters of blood vessels are known to regulate blood flow to the organs [27]. Chu et al. suggested that the VDI could be a parameter to systemically evaluate the change in the retinal vascular diameter [15]. However, the VDI does not reflect the change in the diameter of a single blood vessel. Instead, it is the averaged blood vessel diameter within a specific ROI. Nonetheless, the VDI could help to indirectly investigate how much the blood vessel diameters systemically change between the daily OCT angiograms. Therefore, it could also be one of the additional parameters to help understand the characteristics of an abnormal blood vessel such as a tumor vascular network.

Remarkably, the whole trend in the FD for all the cases was well consistent with that from previous reports by Gazit et al. using an intravital microscope [13, 25]. To study the FD, they used two kinds of tumor cell lines. One was the colorectal adenocarcinoma LS174T cell line for tumor growth, and the other was the androgen-dependent Shionogi murine mammary carcinoma (ASMMC) cell line for both tumor growth and its regression. As both of the tumor lines grew, the trend for the FD went up and approached the steady state which was the same as the plateau for the untreated tumor case in this study (Fig. 6(b)). They performed an orchiectomy to treat tumor progression caused by ASMMC. After the orchiectomy was performed, the FD went down due to the regression of the tumor which was similar to that for the treated tumor case in this study (Fig. 6(a)). Therefore, the FD could be a reliable indicator to describe the change in the tumor blood vessel morphology during tumor growth and its response to treatment.

Lacunarity reflects how well the tumor vascular distribution looks homogeneous. Potentially, it would be a surrogate indicator of the tortuosity to estimate how well chemotherapeutic drug can be transported or gas exchange occurs. Nonetheless, lacunarity as well as VDI need to be analyzed more carefully to be used as biomarkers. This should be performed in further work.

Meanwhile, this study has a few limitations. First, our OCTA system does not have a high spatial resolution, and thus, it could not visualize microvessels in the tumor vasculature. Optical coherence microscopy using a high NA objective lens and tube lens pairs would provide a much higher resolution to visualize down to the capillary level which is comparable to multi-photon microscopy [14]. In addition, over an eight times higher sampling density and fast repeated scanning protocol at the same C-scan position would also enhance the decorrelation signal produced by the perfused blood flow, especially in a capillary compartment [28].

Second, this study was a pilot study to investigate which vascular morphometric parameters correlate well with both tumor growth and tumor response to chemotherapy. Each case had only one animal as a subject. Therefore, it should not be concluded that the VSD and FD correlate with tumor volume significantly better than that of the VDI and lacunarity. Further studies that include a sufficient number of animals for each group will help support our conclusion. However, it is worth noting that the VSD and FD may predict the tumor response earlier than the change in tumor volume after chemotherapy.

Recently, several groups have tried to extract the absolute blood flow velocity using some of the OCTA algorithms (i.e., SSADA [29], MUSIC-OCT [30], and IBDV [24]). If an optimized scanning protocol (i.e., A-scan rate and sampling density) can find an absolute linear velocity range, it will help to quantitatively analyze the tumor blood flow. Therefore, for further study, the measurement of an absolute velocity by OCTA will help to investigate the relationship between tumor blood flow and its response to treatment with an external intervention such as vasoactive agents [31] or inhaled gas modulation [26, 32].

In conclusion, we showed that the changes in some vascular morphometric parameters (i.e., vessel skeleton density, vessel diameter index, fractal dimension, and lacunarity) have positive or negative correlations with the tumor volume during tumor growth and its chemotherapy. Among those parameters, the changes in the VSD and FD occurred one day earlier than the change in the tumor size compared to the changes in the VDI and lacunarity. The result also showed that there is an asymmetric change in the VSD during the time between tumor growth and its regression for the treated case. This feasibility study shows that OCTA morphometrics would be helpful to evaluate the tumor response to treatment and to understand tumor heterogeneity in preclinical studies comprehensively.

Figure A1 shows the selected representative OCT angiograms (OCTA) for a total of fifteen days for the no tumor case and its corresponding color photos. We selected a FOV of 5 mm × 5 mm (as indicated by the white dotted square in the color photos) to take the OCT angiogram. Within the two user-defined ROIs (in the 10 and 5 o’clock directions as indicated by the red dotted areas on the OCTA), the summed value was calculated for both the VSD and VDI while the average values were computed for both the FD and lacunarity. There is an explicit change in the daily OCTA map even for the no tumor case. For example, new blood vessels started to grow until Day 5, and after that, the perfused blood flow tended to decrease. On Day 15, the upper skin got partially torn; thus, the experiment was ended.

Figure A2 shows the daily OCT angiograms and their corresponding color photos from a smartphone for the treated tumor case. The experiment was ended on Day 18 because the upper skin got partially torn as it did in the no tumor case. Nonetheless, the daily characteristic blood vessels within the ROI maintained a similar pattern even though their sizes had changed somewhat. Interestingly, a few of the new blood vessels were formed horizontally as shown in the OCT angiograms from Day 6 to Day 10 while most of the blood vessels within the ROI were destructed on Day 12 (after 4th CTX).

Figure A3 shows the daily OCT angiograms and their corresponding color photos from a smartphone for the untreated tumor case. Interestingly, angiogenesis occurred along the horizontal direction until Day 9 just like in the treated tumor case. Starting from Day 11, however, the horizontally oriented blood vessel disappeared. Instead, the vertically oriented blood vessel was predominant, and the blood vessels started to be destructed as shown by several small black areas in the angiogram which corresponds to a necrotic or apoptotic region.