1.

Introduction

Diffuse optical spectroscopic imaging (DOSI) is a functional imaging technology that measures hemodynamics and biochemical composition in centimeter-thick tissues.¹ We have developed a broadband handheld scanning probe for imaging breast tumors under clinically relevant conditions, such as detecting tumors in mammographically dense tissue,² distinguishing between malignant and benign tumors,³ monitoring tumor response^4–6 to chemotherapy, and assessing breast density.⁷ In these studies, broadband NIR data (650 to 1000 nm) are used to calculate tissue absorption and scattering spectra and develop various metrics of endogenous contrast based primarily on tissue concentrations of oxy- and deoxyhemoglobin (${\mathrm{HbO}}_2$, HbR), water, and lipid. To enhance this information and provide a more complete picture of biochemical composition and oxygen consumption inside and in the vicinity of breast lesions, DOSI can be combined with diffuse correlation spectroscopy (DCS), a complementary method that measures deep tissue blood flow.

DCS utilizes the fluctuations of detected light intensity from the movement of particles to probe microvasculature blood flow in deep tissue.^8–10 The blood flow index (BFI) measured by DCS has been validated by arterial spin-labeled magnetic resonance imaging,¹¹ Doppler ultrasound,^12,13 Xenon computed tomography,¹⁴ and laser Doppler flowmetry.^15,16 Further, DCS systems have been used to monitor blood flow changes in clinical applications associated with brain,^13,14,17 skeletal muscle,^11,18,19 and tumors.^10,20–23 When taken together, DCS and DOSI can provide a noninvasive measurement of the metabolic rate of oxygen consumption in tissue, and indeed, several studies have shown that the combination of DOSI and DCS techniques in breast lesion characterization provides a better understanding of tumor physiology and metabolism and may enhance treatment monitoring in the presurgical setting.^20,23,24 However, these breast lesion studies were performed sequentially by using separate DCS and DOSI instruments. Serial measurements add additional time for data collection and can lead to uncertainties in the origins of the different signals.

This study aims to develop and validate a spatially coregistered, DOSI/DCS metabolic imaging system for enhancing breast tumor characterization. We combined DCS and DOSI techniques into a single handheld probe and developed integrated software that collects the data nearly simultaneously. Since the BFI depends on tissue optical properties, the BFI is typically calculated using fixed values from literature or using values measured with a separate instrument. DOSI is unique because it provides these optical properties (absorption, ${\mu }_{\mathrm{a}}$, and reduced scattering, ${\mu }_{\mathrm{s}}^{\prime }$, parameters) over a broad spectral range from 650 to 1000 nm.²⁵ The impact of ${\mu }_{\mathrm{a}}$ and ${\mu }_{\mathrm{s}}^{\prime }$ on BFI has been investigated in a previous study using a homogeneous liquid phantom where particles undergo Brownian motion.²⁶ We have extended this work to more realistic tissue simulating phantoms that employ a buried flow channel designed to simulate a subsurface flowing inhomogeneity. This allows us to analyze, for the first time, the combined impact of optical properties (${\mu }_{\mathrm{a}}$, ${\mu }_{\mathrm{s}}^{\prime }$), flow velocity, and channel depth on the estimation of the BFI.

Coregistered DOSI and DCS were then applied to construct two-dimensional (2-D) maps of subsurface particle flow in a solid tissue-simulating phantom with an embedded flow channel. Finally, images of tissue BFI, composition, and relative metabolism were obtained, for the first time, by coregistered DOSI/DCS mapping in a clinical measurement of a breast cancer patient. DOSI-derived optical properties at each measurement point were used to correct BFI in both clinical and tissue phantom images, improving accuracy and revealing unique and complementary contrast in breast cancer.

2.

Material and Methods

2.1.

DCS Theory

A typical DCS setup using optical fibers in tissue contact for light source and detector is shown in Fig. 1(a). The light source utilizes a long coherence length laser. The backscattered light intensity $I(t)$ is detected at a distance $\rho$ from the source. The temporal speckle fluctuations of $I(t)$ is analyzed by calculating the normalized light intensity temporal autocorrelation function ($g_2$): $g_2(\tau )=\frac{⟨I(t)I(t+\tau )⟩}{{⟨I(t)⟩}^2}$, where $\tau$ represents the correlation delay time. Temporal speckle fluctuations cause variations in detected light intensity; thus, motions of red blood cells in microvascular blood vessels (arterioles, capillaries, and venules) provide the dominant contrast mechanism.^8,11,27,28 Speckle fluctuations from a medium with higher flow rate produce a faster decay of $g_2$. The Siegert equation²⁹ relates the normalized electric field autotemporal correlation function, $g_1(\tau )$, to $g_2(\tau )$

Eq. (1)

$$g_2(\tau )=1+\beta {|g_1(\tau )|}^2,$$

where $\beta$ depends on the optical setup and can be experimentally determined.³⁰

Fig. 1

(a) The semi-infinite measurement geometry for DCS including a positive isotropic light sources and a negative isotropic light (imaging) source for DCS theory; see Sec. 2.1 and Ref. 30 for details. (b) Block diagram of the DOSI/DCS instrument, red color represents the multimode source fibers. Collection fiber bundle is shown by black, which contains a DOSI spectrometer fiber and two DCS single mode fibers (shown with yellow), and orange is an RF cable. (c) Photograph of the entire system on a portable cart and (d) diagram and photograph of handheld probe.

Similar to the photon diffusion equation, a correlation diffusion equation that describes the propagation of the unnormalized electric field temporal autocorrelation function $G_1(\tau )$ in a turbid media can be derived. For the setup in Fig. 1(a),^{8,9,15,30–33}

Eq. (2)

$$G_1(\rho ,z,\tau )=\frac{3{\mu }_{\mathrm{s}}^{\prime }}{4\pi }[\frac{e^{-k(\tau )r_1}}{r_1}-\frac{e^{-k(\tau )r_2}}{r_2}],$$

where $r_1={[{\rho }^2+{(z-z_0)}^2]}^{\frac{1}{2}}$, $r_2={[{\rho }^2+{(z+z_0+2z_b)}^2]}^{\frac{1}{2}}$, $z_0=\frac{1}{{\mu }_{\mathrm{s}}^{\prime }}$, $z_b=\frac{2(1+R_{\mathrm{eff}})}{3{\mu }_{\mathrm{s}}^{\prime }(1-R_{\mathrm{eff}})}$, and $R_{\mathrm{eff}}=-1.440n^{-2}+0.71n^{-1}+0.668+0.0636n$ is the effective reflection coefficient, which is determined by the ratio of the refraction indices of two medium (e.g., $n=\frac{n_{\text{tissue}}}{n_{\text{air}}}\approx 1.33$). Equation (2) incorporates the extrapolated zero boundary condition, which approximates the isotropic light source located at $z=z_0$ and negative isotropic imaging source located at $z=-(z_0+2z_b)$. Schematic of the extrapolated boundary geometry is shown in Fig. 1(a).^30,32–34 The effective decay rate $k(\tau )$ of $G_1(\rho ,\tau )$ provides information about the motion of the scatterers in the underlying medium: $k^2(\tau )=3{\mu }_{\mathrm{s}}^{\prime }{\mu }_{\mathrm{a}}+{\mu }_{\mathrm{s}}^{\prime 2}{k}_0^2\alpha ⟨\mathrm{\Delta }{r}^2(\tau )⟩$ where $⟨\mathrm{\Delta }{r}^2(\tau )⟩$ is the mean square displacement of the moving particles, $k_0$ is the wave-vector magnitude of the incident light field, and $\alpha$ is proportional to the fraction of scatterers that are moving (i.e., the fraction of tissue scatterers that are red blood cells). For diffusive motion $⟨\mathrm{\Delta }{r}^2(\tau )⟩=6D_b\tau$, where $D_b$ is the effective Brownian diffusion coefficient for the tissue scatterers.³⁰ The BFI (${\mathrm{mm}}^2/\mathrm{s}$) is defined as $\mathrm{BFI}=\alpha D_b$.^15,35 We used the tissue optical properties measured with DOSI at each measurement iteration to calculate BFI. Assuming $\alpha =1$, the factor $k(\tau )$ is fitted to the measured autocorrelation function, $g_2(\rho ,\tau )$, via Eqs. (1) and (2) to derive the BFI.¹⁵

3.

Results

3.1.

Liquid Phantom Characterization

The measured intensity autocorrelation functions, $g_2$, for different flow rates are shown in Fig. 2(a). The $g_2$ exhibits faster decay in the presence of higher flow rate.

Fig. 2

DCS characterization. (a) Measured intensity autocorrelation functions $g_2$, during liquid phantom depth/flow experiment for flow tube 5 mm below the surface. (b) The relative change of blood flow, rBF, versus pump flow rates, for various flow tube depths. Depth is the distance between the top of the flow tube and the handheld probe. Data are the average of 10 samples, and error bars indicate standard deviation. The DOS-measured optical properties of the liquid phantom were ${\mu }_{\mathrm{s}}^{\prime }=0.75{\mathrm{mm}}^{-1}$ and ${\mu }_{\mathrm{a}}=0.005{\mathrm{mm}}^{-1}$ at 785 nm.

Figure 2(b) shows the effect of flow tube depth on rBF for different flow rates. BFI was calculated using the corresponding ${\mu }_{\mathrm{a}}$ and ${\mu }_{\mathrm{s}}^{\prime }$ values ($0.0048{\mathrm{mm}}^{-1}$ and 0.78 mm, respectively) from DOSI. Compared to a superficial flow tube, the BFI decreased to 52%, 24%, 10%, and 8% (mean ratio over all flow rates) when the depth of the tube was 5, 10, 15, and 20 mm, respectively.

To illustrate the effect of absorption error on the BFI, two groups of BFI were calculated while varying the background IL mixture ${\mu }_{\mathrm{a}}$ with a constant flow rate. The first group was calculated by assuming constant ${\mu }_{\mathrm{a}}=0.0093{\mathrm{mm}}^{-1}$ (median value), and the second group was calculated by using the extracted ${\mu }_{\mathrm{a}}$ value from DOSI at 785 nm in each step. The average BFI for each group is shown in Fig. 3(a). BFI measured with constant ${\mu }_{\mathrm{a}}$ value dropped slightly for a constant flow rate during the absorption titration. However, BFI recovered using DOSI-derived values remained nearly constant (with less than 5% variation). We observed that a 100% increase in ${\mu }_{\mathrm{a}}$ (from 0.0076 to $0.015{\mathrm{mm}}^{-1}$) induced almost 20% range of error on reconstructed BFI. The DOSI-measured ${\mu }_{\mathrm{s}}^{\prime }$ ($\mathrm{M}\pm \mathrm{SD}$) over the absorption titration was $0.74\pm 0.031{\mathrm{mm}}^{-1}$. Increased variation in BFI was noted at higher absorption cases due to the decreased DCS count rate and signal-to-noise ratio.

Fig. 3

Effect of absorption and scattering on BFI. BFI calculated with DOSI-corrected and constant optical properties are shown for (a) constant scattering and varied absorption and (b) constant absorption and varied scattering. The flow rate was constant during the measurements.

Next, the background IL concentration and thus ${\mu }_{\mathrm{s}}^{\prime }$ were varied with a constant flow tube depth and rate, with a constant absorption ($\mathrm{M}\pm \mathrm{SD}$) of $0.0041\pm 0.00027{\mathrm{mm}}^{-1}$. Similar to the absorption test, two groups of BFIs were measured either by assuming ${\mu }_{\mathrm{s}}^{\prime }$ was constant ($0.8{\mathrm{mm}}^{-1}$) or by using the DOSI-measured ${\mu }_{\mathrm{s}}^{\prime }$. Illustrated in Fig. 3(b), a 60% increase of ${\mu }_{\mathrm{s}}^{\prime }$ (from 0.59 to $0.94{\mathrm{mm}}^{-1}$) caused a 53% range of error in BFI—significantly more than the same proportional change in absorption. When using DOSI-correction, BFI was more consistent (with less than 12% variation). In general, we observed that an uncertainty of $0.3{\mathrm{mm}}^{-1}$ in ${\mu }_{\mathrm{s}}^{\prime }$ leads to $\sim 53\%$ error in BFI, while an uncertainty of $0.003{\mathrm{mm}}^{-1}$ in ${\mu }_{\mathrm{a}}$ yields only 9% error.

3.2.

Silicone Phantom

A phantom with an embedded flow channel was made to demonstrate imaging with the combined probe. The measured optical properties of the phantom (outside of the flow channel) were ${\mu }_{\mathrm{s}}^{\prime }=0.8{\mathrm{mm}}^{-1}$ and ${\mu }_{\mathrm{a}}=0.008{\mathrm{mm}}^{-1}$ at 785 nm. DOSI and DCS images calculated from two scans (no flow and $200\mathrm{mL}/\min$ flow) are shown in Fig. 4. The values of ${\mu }_{\mathrm{s}}^{\prime }$ and ${\mu }_{\mathrm{a}}$ ($\mathrm{M}\pm \mathrm{SD}$) for the three vertical lines (total of 19 before interpolation) centered in the tube were $0.88\pm 0.0098{\mathrm{mm}}^{-1}$ and $0.010\pm 0.00097{\mathrm{mm}}^{-1}$, respectively. BFI values ($\mathrm{M}\pm \mathrm{SD}$) in this area were $4.42\times {10}^{-6}\pm 2.52\times {10}^{-6}{\mathrm{mm}}^2/\mathrm{s}$ for flow $200\mathrm{mL}/\min$ and $3.22\times {10}^{-6}\pm 3.19\times {10}^{-7}{\mathrm{mm}}^2/\mathrm{s}$ for $\text{flow}=0\mathrm{mL}/\min$.

Fig. 4

DOSI and DCS images of silicone phantom with an embedded flow channel. (a, b) Absorption coefficient ${\mu }_{\mathrm{a}}$, (c, d) reduced scattering coefficient ${\mu }_{\mathrm{s}}^{\prime }$, and (e, f) BFI. Images on left represent no flow (a, c, and e), where the images on the right represent, flow (b, d, and f). Tick mark separation equals 5 mm.

3.3.

Human Breast Tumor Imaging

DOSI and DCS images were acquired on a human breast tumor in vivo. The DOSI-measured optical properties over the tumor and surrounding normal tissue are shown in Figs. 5(a) and 5(b). DCS data were collected in a smaller region over the tumor, and images of BFI were calculated using DOSI measured optical properties [Fig. 5(c)] and using constant normal premenopausal breast optical property values (${\mu }_{\mathrm{a}}=0.005{\mathrm{mm}}^{-1}$ and ${\mu }_{\mathrm{s}}^{\prime }=0.8{\mathrm{mm}}^{-1}$) used in a prior DCS study²⁰ [Fig. 5(d)]. Peak contrasts in these maps [Figs. 5(c) and 5(d)] were measured by normalizing the max BFI on the tumor region to the BFI over a normal region (average of three BFI values far from the tumor). The BFI reconstructed with DOSI measured optical properties had a 50% higher peak contrast than BFI reconstructed using constant optical properties values (267% versus 217%). Figures 6(a)–6(e) show images of breast tissue chromophore concentrations (${\mathrm{HbO}}_2$, HbR, water, lipid, and TOI) calculated from DOSI-measured absorption. ${\mathrm{rMMRO}}_{2(\mathrm{T}/\mathrm{N})}$ calculated from the DOSI-measured ${\mathrm{HbO}}_2$ and HbR and the BFI is shown in Fig. 6(f).

Fig. 5

Integrated DOSI and DCS imaging during the clinical measurement of a breast cancer patient (a) ${\mu }_{\mathrm{s}}^{\prime }$ and (b) ${\mu }_{\mathrm{a}}$ ${\mathrm{mm}}^{-1}$ at 785 nm, (c) BFI (${\mathrm{mm}}^2/\mathrm{s}$) reconstructed with ${\mu }_{\mathrm{s}}^{\prime }$ and ${\mu }_{\mathrm{a}}$ values measured by DOSI at 785 nm in each, (d) BFI (${\mathrm{mm}}^2/\mathrm{s}$) calculated with constant ${\mu }_{\mathrm{s}}^{\prime }$ and ${\mu }_{\mathrm{a}}$ values (0.8 and $0.005{\mathrm{mm}}^{-1}$, respectively). Dashed lines (a and b) represent the region where the lesion was located and DOSI/DCS data were taken. The areola region is outlined with a semicircle. Tick marks represent 10 mm separations.

Fig. 6

Integrated DOSI and DCS imaging during the clinical measurement of a breast cancer patient. (a) ${\mathrm{HbO}}_2$ ($\mu \mathrm{M}$), (b) HbR ($\mu \mathrm{M}$), (c) water (%), (d) lipid (%), (e) tissue optical index ($\mathrm{TOI}=[\mathrm{HbR}]\times [\text{water}]/[\text{lipid}]$), (f) ${\mathrm{rMMRO}}_2$ ($T/N$), and (g) BFI (${\mathrm{mm}}^2/\mathrm{s}$). Dashed lines (a–e) represent the region where the lesion was located and DOSI/DCS data were taken. The areola region is outlined with a semicircle. (f and g) The region that DOSI/DCS data were collected. Tick marks represent 10 mm separations.

For convenient comparison, the BFI from the same region is shown in Fig. 6(g). We observed that the spatial distributions of BFI and ${\mathrm{rMMRO}}_{2(T/N)}$ around the tumor share similar features. However, the spatial distributions of TOI and ${\mathrm{rMMRO}}_{2(\mathrm{T}/\mathrm{N})}$ do not match, but the contrast from the areola is apparent in both images.

4.

Discussion

In this study, we fabricated a combined DOSI/DCS metabolic imaging system designed specifically for breast tissue characterization, validated performance in several phantom studies, and performed breast cancer measurements in a human subject. A single integrated handheld probe was used to collect coregistered DOSI and DCS data from nearly identical tissue regions. Usability and data collection time were optimized by creating a custom software package that allowed the operator to enable either or both optical techniques.

The system was characterized and validated in several liquid and solid phantom experiments. First, a liquid phantom system with a submerged flow tube was developed to evaluate the impact of tube depth, flow speed, and background liquid optical properties on BFI. To simulate the DCS signals relevant to breast tissue, flow rates were selected to yield rBF (tumor/normal) values (from 1.90 to 2.70) at a tube located 15 mm below the surface based on previously measured flow changes of 32 patients.²⁰ Increasing the depth of the flow tube not only decreases the measured flow rate but also reduces flow rate sensitivity and dynamic range. We found that with 28 mm s-d separation, a linear flow signal was accessible up to 15 mm deep. However, while flow changes were accessible for a 20-mm deep flow tube, its linear dynamic range was greatly reduced. $R^2$ values derived from linear regression analysis of BFI at depths of 0, 5, 10, 15, and 20 mm were 0.98, 0.98, 0.98, 0.86, and 0.76, respectively.

Next, we separately varied the ${\mu }_{\mathrm{a}}$ and ${\mu }_{\mathrm{s}}^{\prime }$ of the background liquid to demonstrate how bulk tissue optical properties can affect the DCS assessment of BFI. In previously published work,²⁶ a homogeneous liquid phantom was used to illustrate the effect of ${\mu }_{\mathrm{a}}$ and ${\mu }_{\mathrm{s}}^{\prime }$ on BFI by comparing these values with Brownian motion of particles. In this work, we examined the impact of ${\mu }_{\mathrm{a}}$ and ${\mu }_{\mathrm{s}}^{\prime }$ on BFI using a liquid phantom system with a submerged flow tube. Our experiments show that the BFI error is significantly higher for inaccuracies in ${\mu }_{\mathrm{s}}^{\prime }$ compared to ${\mu }_{\mathrm{a}}$, and it is independent of the flow rate. Our results are comparable to those from a previous study that showed an uncertainty of $0.3{\mathrm{mm}}^{-1}$ in reduced scattering leads to $\sim 50\%$ error in BFI, while an uncertainty of $0.003{\mathrm{mm}}^{-1}$ in absorption yields less than 10% error in BFI at 785 nm.²⁶ Although our absorption and scattering values were chosen to simulate breast tissue, they were varied over a large enough range to provide insight on DCS performance in other tissue types such as muscle and brain. Several phantom studies have investigated BFI error in cerebral blood flow measurements,^38,39 and they have improved BFI accuracy through more complex modeling that accounts for irregular tissue geometries.^40,41 It is possible that BFI accuracy in breast measurements can be further improved using models that account for arbitrary geometries⁴² and tissue heterogeneity.⁴³ These factors should be addressed in future studies.

A solid tissue-simulating silicone phantom with an embedded flow channel was fabricated for the first time to demonstrate the spatially coregistered information content of ${\mu }_{\mathrm{a}}$, ${\mu }_{\mathrm{s}}^{\prime }$, and BFI. In the region over the flow tube, we observed the expected increase in contrast in absorption, scattering, and BFI in both flow and no-flow conditions. BFI contrast in the no-flow case was likely due to the Brownian motion of the liquid scatterers. From these measurements, it is clear that the system is mainly sensitive to deep tissue lying within the field of view defined by the source–detector separation. Limited information is obtained from deep tissue more than 5 mm laterally outside the source–detector plane. Interestingly, the maximum absorption contrast was observed when either the source or detector was positioned directly over the phantom, while the maximum scattering contrast occurred at the midpoint of the source and detector, observations that were confirmed by Monte Carlo simulations (data not presented). Finally, even though the phantom was designed with a centered flow tube, we note that the flow channel appears shifted to the left in all the images. We verified that this is because the reference point used to position the probe was not located at the exact midpoint between the optical sources and detectors.

The combined DOSI/DCS system was used to characterize a high-grade breast tumor in a young patient. Two different images of BFI were calculated by using constant ${\mu }_{\mathrm{a}}$ and ${\mu }_{\mathrm{s}}^{\prime }$ values and by extracting these values from the coregistered DOSI data. Although both BFI images showed increased blood flow in the tumor region (slightly different spatial variation), the BFI images reconstructed with DOSI ${\mu }_{\mathrm{a}}$ and ${\mu }_{\mathrm{s}}^{\prime }$ exhibited higher BFI. This is due to overestimation of ${\mu }_{\mathrm{s}}^{\prime }$ (without DOSI) in some tissue regions leading to an underestimation of BFI. Overall, DOSI derived ${\mu }_{\mathrm{a}}$ and ${\mu }_{\mathrm{s}}^{\prime }$ can vary up to 100% and 60% over different tissue regions, respectively, and the maximum error in the BFI reconstructed with constant ${\mu }_{\mathrm{a}}$ and ${\mu }_{\mathrm{s}}^{\prime }$ values ranges from 20% to 53%. These findings underscore the importance of determining tissue optical and DCS properties from the same location. A prior study indicated that fixed optical properties did not significantly affect the relative change of BFI in tumor versus normal regions in a group of cancer patients. However, individual relative changes in BFI were clearly impacted when using DOSI-measured optical properties.²⁰ In addition, assuming constant optical properties can be particularly impactful when monitoring a subject during chemotherapy, since tumor optical properties are changing differently from the background.⁵

Our clinical data further underscore the importance of incorporating blood flow in characterizing breast tumors. Figure 6(f) shows the spatial distribution of oxygen metabolism in and around the tumor compared to the TOI, a high-contrast function used to localize tumors.²⁹ We observe colocalized contrast in the images due to the metabolically active tissue of the areola. However, both the peak locations and distributions of TOI and ${\mathrm{rMMRO}}_{2(T/N)}$ around the tumor do not match. For example, one region of the tumor indicated by increased TOI contrast [$x$, $y=2$ to 3 cm in Fig. 6(e)] corresponds to a relatively low BFI and ${\mathrm{rMMRO}}_{2(T/N)}$. This is consistent with an increased vascular density but a lower blood flow that results in poor tissue perfusion in this region of the tumor. Consequently, there is a region adjacent to the peak TOI with a high BFI and ${\mathrm{rMMRO}}_{2(T/N)}$, which may be due to the increased perfusion of peripheral tumor tissue. Overall, these data show how blood flow information augments and complements compositional data to provide a more complete picture of tumor perfusion and metabolism.

5.

Conclusion

We describe a clinically compatible optical imaging system that simultaneously collects tissue composition and blood flow data using DOSI and DCS, respectively. Specifically, the system measures tissue concentrations of oxy- and deoxyhemoglobin, water, and lipid, as well as blood flow. We validated instrument performance in both liquid and solid phantom systems and demonstrated coregistered imaging in phantoms. Measurements in a patient with breast cancer confirm the importance of accounting for spatial variations in tumor ($T$) and normal ($N$) tissue optical properties for optimizing BFI sensitivity and contrast. Finally, hemoglobin and flow parameters were combined to create images of the $T/N$ relative metabolic rate of oxygen consumption [${\mathrm{rMMRO}}_2$ ($T/N$)]. Together, quantitative DOSI and DCS imaging provides a more complete view of tumor composition and metabolism, which may aid in predicting, guiding, and personalizing cancer therapies.