2.1.

Scatter Imaging System

The scatter imaging system consists primarily of a confocal spectroscopic system having $100\text{-}\mu \mathrm{m}$ illumination spot size and a raster-scanning sample platform built using linear translation stages. The optical and electromechanical subsystems are integrated via a custom-developed LabVIEW (National Instruments, Austin, Texas) interface. Figure 1 shows the schematic of this system, which resembles a traditional confocal microscope but is designed to directly image scatter from tissue. This is achieved by ensuring that the illumination and detection spot sizes on the tissue surface are less than one scattering length (typically $100\mu \mathrm{m}$ for tissue)²⁷ over the entire waveband of interest. In this regime, the incident photons are predominantly singly scattered and the detected signal is largely independent of local absorption.

Fig. 1

Schematic of the raster-scanning scatter imager. The solid semitransparent arrows show the illumination and detection beam paths. Identical achromatic lenses (L1 and L3), the objective lens (L2), the beamsplitter (BS), the illumination fiber (F1), and the detection fiber (F2) are all arranged in the standard confocal geometry. The distal end of the detection fiber is coupled to a CCD-based imaging spectrometer (SPEC), which records the remitted irradiance spectrum.

The illumination optics train consisted of a $50\text{-}\mu \mathrm{m}$ core fiber (F1) coupled to a $20\text{-}\mathrm{W}$ tungsten-halogen white light source (HL-2000-HP, Ocean Optics, Inc., Dunedin, Florida) placed at the front focal point of an achromatic lens (L1) (PAC040, Newport Corporation, Irvine, California). A $10\times$ , 0.28 NA, long working distance, air immersion, plan-apochromatic objective (L2) was used to refocus the light on to the sample plane. The illumination optics train was modeled in ZEMAX optical design software (version July 22, 2004, ZEMAX Development Corporation, Bellevue, Washington) to make sure that the illumination spot size was less than $100\mu \mathrm{m}$ over the entire wavelength band used. The $50\text{-}\mu \mathrm{m}$ -diam extended source (fiber core) was modeled as a collection of point sources placed at the extremities of the source field positions. The transverse magnification of the illumination optics train was approximately 0.4. The theoretical spot size at the sample plane was found to be less than approximately $40\mu \mathrm{m}$ over a broad wavelength range of $450\mathrm{nm}\text{to}800\mathrm{nm}$ , with chromatic aberration dominating the point spread function. The theoretical spot size was within $100\mu \mathrm{m}$ for a depth of approximately $130\mu \mathrm{m}$ on either side of the best focus plane. It should be noted that the microscope objective (L2) lens system was modeled as a single paraxial lens with equivalent specifications, as the actual lens data for the objective was not available. However, the objective is well corrected for common optical aberrations and was expected to deliver a diffraction-limited performance over the design waveband. The spot size was also experimentally measured using the standard knife edge technique and was found to be approximately $89\mu \mathrm{m}$ .

The detection optics train used the same microscope objective to pick up the backscattered light from the sample, and a $50∕50$ beam splitter (BS) was used to separate the illumination and detection beam paths. Another achromatic lens (L3) was used to focus the detected photons onto the proximal face of a $100\text{-}\mu \mathrm{m}$ -core detection optical fiber (F2), which also acted as the confocal pinhole. The size of the detection spot on the target was controlled by the detection fiber’s core diameter and the lateral magnification of the optical system. The detection optical train provided a magnification of approximately 2.5, and thus the $100\text{-}\mu \mathrm{m}$ detection fiber picked up light from a $40\text{-}\mu \mathrm{m}$ region overlapping the illumination spot on the sample plane. The distal end of the detection fiber is coupled to an imaging spectrometer (SpectraPro 2156, Princeton Instruments, Inc., Trenton, New Jersey) with a $150\text{grooves}∕\mathrm{mm}$ , $500\text{-}\mathrm{nm}$ blaze wavelength, ruled grating and a $16\text{-}\text{bit}$ , thermo-electrically cooled Cascade 512F EM-CCD (Princeton Instruments, Inc.). The CCD has a field of $512\times 512\text{pixels}$ with a pixel size of $16\text{microns}$ . The spectrometer was calibrated using characteristic spectral lines from a calibrated fluorescent lamp PLS 11W/827 (Philips Lighting, Somerset, New Jersey)³¹ to operate in the wavelength range of $510\mathrm{nm}\text{to}785\mathrm{nm}$ that encompasses the strong hemoglobin absorption peaks. The spectral resolution was approximately $1\mathrm{nm}$ .

A custom sample mount was designed to hold removable microscope slides and standard 96-well plates for tissue and liquid phantom measurements, respectively. The mount was integrated into an XY linear translation platform and aligned such that the top surface of the microscope slide lies exactly on the focal plane of the objective lens (L2). The entire optical head was mounted on a vertical translation stage to enable the scan spot to move in and out of the sample plane to allow for precise vertical alignment.

At normal incidence, Fresnel reflections off the glass-sample interface can cause significant measurement artifacts, so the sample plane was tilted by about $45\mathrm{deg}$ with respect to the optical axis to prevent specular reflections from entering the detection path. A significant portion of the illumination beam still coupled into the detection path and was found to be due to Fresnel reflection from one of the beamsplitter faces. This signal, however, remained constant during the measurement process and was mathematically removed by measuring a background spectrum, $I_{\mathit{bg}}\left(\lambda \right)$ , with no sample present. A block of Spectralon (Labsphere, Inc., North Sutton, New Hampshire), a spectrally flat thermoplastic resin, was used to collect a reference spectrum, $I_{\mathit{ref}}\left(\lambda \right)$ . At each pixel location, the remitted scatter spectrum $\left(I_R\right)$ was calculated by dividing the measured spectrum $\left(I_{\mathit{meas}}\right)$ by the reference spectrum $\left(I_{\mathit{ref}}\right)$ to remove the instrumental spectral response, as shown in the equation:

2.2.

Parameter Fitting

In the absence of significant local absorption, the relationship between the measured irradiance and wavelength would be approximated by a power law–type empirical relation:

The spatial extent of illumination and detection was less than one mean free path length of interaction (typically $100\mu \mathrm{m}$ ), so the detected photons are thought to be predominantly singly scattered, and the measured spectrum is independent of local absorption. However, in regions where high local concentration of chromophores is encountered, like regions with blood pooling or high vascularity, this approximation is not valid, and a correction for absorption is required to describe the scatter spectrum accurately. The exponential term in Eq. 3 serves as this correction factor, with the empirical fitting process providing an estimate on the product of the path length and the whole blood concentration, $kc$ and the hemoglobin oxygen saturation fraction $d$ . It should be noted that, in this probe geometry, it is not possible to decouple the path length $k$ and the absolute chromophore concentration $c$ from the measured spectra.

Another parameter of interest is the average scattered irradiance, $I_{\mathit{avg}}$ , integrated over all wavelengths and given mathematically by:

2.3.

Animals

All animals were housed in the animal facility with $12\text{-}\mathrm{h}$ light–dark cycles, with temperature control, free access to water, and standard laboratory diet. All procedures and experiments were approved by the Animal Care and Use Committee of Dartmouth College.

2.4.

Prostate Tumors

Orthotopic prostate tumors were grown in male Copenhagen rats (average $120\text{to}150\mathrm{g}$ body weight, Charles River Laboratories, Wilmington, Massachusetts).³³ For all procedures, the rats were anesthetized with an intramuscular injection of a mixture of ketamine and xylazine $(90:9\mathrm{mg}∕\mathrm{kg})$ . For orthotopic implants, the rat abdominal area and hind flanks were shaved and prepared in an aseptic surgical manner. R3327 MatLyLu Dunning prostate cancer cells³⁴ were cultured in RPMI 1640 with glutamine (Mediatech, Herndon, Virginia) supplemented with 10% fetal bovine serum (FBS, HyClone, Logan, Utah) and $100\text{units}∕\mathrm{mL}$ penicillin-streptomycin (Mediatech, Herndon, Virginia). The tumor was induced by injection of $1\times {10}^5$ cells on the right flank in $0.05\text{-}\mathrm{mL}$ sterile phosphate buffered saline (PBS). Tumor growth was assessed by external measurements with calipers, three times per week. Tumors were used for the experiment at $9\text{to}12\text{days}$ after inoculation, with a surface diameter of $7\text{to}9\mathrm{mm}$ and a thickness of $2\text{to}4\mathrm{mm}$ .

2.5.

Pancreatic Tumors

For the subcutaneous pancreatic tumors, male C.B.-17 strain 236 SCID mice ( $6\text{to}7\text{weeks}$ , Charles River Laboratories, Wilmington, Massachusetts) were used. Human pancreatic tumor cells, AsPC-1, were grown and maintained in RPMI 1640 with $2\text{-}\mathrm{mM}$ L-glutamine with $25\text{-}\mathrm{mM}$ HEPES (Lonza, Walkersville, Maryland), and $1.0\text{-}\mathrm{mM}$ sodium pyruvate and fetal bovine serum, 10% (Atlanta Biologicals, Lawrenceville, Georgia), with $100\text{units}∕\mathrm{mL}$ of penicillin-streptomycin (Mediatech, Herndon, Virginia). Tumors were induced by injecting $1\times {10}^6$ AsPC-1 cells ( $50\mu \mathrm{L}$ volume) subcutaneously in the flank region. Cells were suspended in a solution of 50% culture medium (without FBS) and 50% Matrigel (BD Biosciences, Bedford, Massachusetts), loaded into insulin syringes, and kept on ice until they were injected under the skin. Tumors were measured (calipers) weekly throughout the seven week study. When harvested for this experiment, the tumors measured $6\text{to}7\mathrm{mm}$ in diameter and $5\text{to}6\mathrm{mm}$ in thickness.

2.6.

Tumor Imaging

Excised whole tumors were dissected into $4\text{to}5\text{-}\mathrm{mm}$ -thick sections and imaged using the scatter imaging system. The scan resolution was maintained at $100\mu \mathrm{m}$ , and the measurements were referenced to a Spectralon standard as before. The sample was held inside a rubber gasket and sandwiched between two glass slides, with the imaged surface facing down. This arrangement kept the sample in a reasonably airtight environment and also ensured that a flat surface was presented to the scanning beam. The scanning sample stage took approximately $1\mathrm{h}$ to scan the entire field of interest. A few drops of PBS were placed on the sample before the start of the scan to prevent it from drying during the measurements. Addition of PBS did not alter the signal in a significant manner. It was assumed that the sample optical properties did not change significantly during the measurement process. After the measurement, the sample was immediately placed in 4% neutral buffered formaldehyde solution for fixation and routinely processed for histology evaluation. All histology sections were superficially cut at $4\mu \mathrm{m}$ and stained with hematoxylin and eosin (H&E) for imaging analysis.

2.7.

Histopathology Classification of Tumor Subregions

A veterinary pathologist (P. J. H.) examined the H&E sections from each sample and identified several regions-of-interest corresponding to the tissue types observed. The tissue types observed across all samples were classified under three major groups with constituent subgroups, as listed in Table 1 . The epithelial group was identified as containing two distinct types of epithelial cells, one that exhibited higher nucleus to cytoplasm ratio compared to the other. As higher nucleus-to-cytoplasm ratio is indicative of cell proliferation, the subgroup containing this type of cells was identified as the high proliferation index (HPI) tumor cells. The subgroup containing cells with comparatively lower nucleus-to-cytoplasm ratio was identified as the low proliferation index (LPI) tumor cells. Regions exhibiting significant fibrosis were classified into early, intermediate, and mature fibrosis subgroups based on the maturity of the tumor’s fibroplastic response. Exudative necrosis, necrosis marked by the presence of exudative fluids, and focal necrosis, spots of dead cells found within a viable tumor region, were the two types of necrotic regions observed in the measured samples.

3.1.

Phantom Measurements

Since the scatter imager employed a confocal detection scheme, fluctuations in the sample plane’s z-position could cause fluctuations in the measured signal. A homogeneous 5% IntraLipid phantom obtained by diluting a stock 20% IntraLipid solution (20% phospholipid-stabilized soybean oil in water) was used to test the spatial uniformity of the scanner over a $1\mathrm{cm}\times 1\mathrm{cm}$ scan area. Figures 2a and 2b show the average scattered irradiance image and its histogram. A batch of phantoms of varying volume concentrations of the stock IntraLipid solution was used to test the linear response of the system. The system response, as shown in Fig. 2c, suggests that in the absence of absorption, the average irradiance of the remitted light is approximately proportional to the reduced scattering coefficient.

Fig. 2

(a) and (b) Average scattered irradiance image of a $1\mathrm{cm}\times 1\mathrm{cm}$ homogeneous 5% IntraLipid phantom and its associated histogram. (c) Plot of average scattered irradiance as a function of IntraLipid concentration. (d) Plots of average scattered irradiance (values scaled by a factor of 0.1) and scatter power as a function of blood concentration in a 1% IntraLipid phantom. The estimated scatter parameters are shown to be relatively independent of the changes in blood concentration. (e) Plot of estimated blood concentration index, $kc$ , as a function of blood concentration. (f) Plot of estimated hemoglobin oxygen saturation fraction, $d$ , in phantoms with different concentration of blood.

To test the probe’s insensitivity to local absorption changes, another batch of IntraLipid phantoms was prepared with varying volume concentrations of whole animal blood. Heparinized whole animal blood harvested from a sheep was used, and the IntraLipid concentration was fixed at 1% ( ${\mu }_a=2.3\times {10}^{-4}{\mathrm{mm}}^{-1}$ and ${\mu }_s^{\prime }=1.1{\mathrm{mm}}^{-1}$ at $633\mathrm{nm}$ , scaled from reported values for IntraLipid 10%).³⁵ The estimated scatter parameters were expected to remain constant across this batch of phantoms despite the change in absorption. As expected, the average scattered irradiance and scattering power remained constant over increasing whole blood concentration, as shown in Fig. 2d. Estimated blood concentration index $\left(kc\right)$ as shown in Fig. 2e was negligible for low concentrations of blood. This behavior at low blood concentrations is expected to be due to the inherent insensitivity of the probe to local absorption. However, for blood concentration values beyond 1%, an approximately linear increase was observed. The estimated hemoglobin oxygen saturation fraction is shown in Fig. 2f. It should be noted that the estimates of $d$ are not reliable in phantoms with low concentration of blood due to the instrument’s insensitivity to absorption under these conditions.

3.2.

Tumor Measurements

Six pancreas and four prostate tumor samples were imaged using the scatter imaging system. The average scattered irradiance image and the scattering power images were extracted from the raw spectral images and are shown in Figs. 3a, 3b, 3c, 3d . Figure 4 shows representative spectra acquired from pancreas and prostate tumor samples along with their corresponding empirical fits. Under the guidance of a veterinary pathologist, several regions-of-interest corresponding to tissue types classified in Table 1 were identified on each of the measured tumor samples. Figure 5 shows an example where five regions-of-interest identified in a pancreatic tumor sample are shown overlaid on the scatter power image along with a close-up view ( $100\times$ magnification) of the corresponding H&E sections. Region 1 shows HPI tumor cells with more cellular density compared to LPI tumor cells found in Region 2. Regions 3 and 4 show early and intermediate stages of fibrosis marked by the increased stromal content and less cellularity. Region 5 shows necrosis with the presence of exudative fluids and fragments of dead cells.

Fig. 3

(a) and (b) Average scattered irradiance and scatter power images of the measured pancreas tumor (AsPC-1) samples. The third image from the left contains two samples that were imaged together. (c) and (d) Average scattered irradiance and scatter power images of the measured prostate tumor (Mat-LyLu) samples.

Fig. 4

Representative reflectance spectra measured in pancreas and prostate tumor samples along with their corresponding empirical fits to Eq. 3 (solid line).

Fig. 5

Scatter power image of a pancreas tumor sample showing five pathology-based regions-of-interest overlaid on it. Variations in tissue ultrastructure are evident on the shown high magnification $(100\times )$ pathology images: (1) High proliferation index (HPI) tumor cells, (2) low proliferation index (LPI) tumor cells, (3) significant early fibrosis with some lipids, (4) necrosis (exudative), and (5) mature fibrosis.

The average scattered irradiance and scatter power data for different tissue types were obtained by sampling corresponding histology-guided regions-of-interest in the scatter images. Figures 6a and 6b show the grouped scatter plots of scattering power versus average scattered irradiance for pancreas and prostate tumor samples. Figures 7a and 7b show the box and whisker plots of average scattered irradiance and scatter power distributions for different tissue types found in the pancreas tumor samples. Figures 8a and 8b show the same for the prostate tumor samples. The boxes in these plots contain the “middle 50%” of the data distribution they represent, bounded by the first and the third quartiles. The line within the boxes represents the median (second quartile). The whiskers extend to 1.5 times the interquartile range from either end of the boxes. The notch around the median lines represents the 95% confidence interval within which the “true median” of the represented distribution exists. If the notches in two plotted distributions do not overlap, then one can conclude with 95% confidence that their true medians differ. The + symbols outside the whiskers indicate the outliers.

Fig. 6

(a) and (b) Grouped scatter plots of scattering power versus average scattered irradiance for all significant tissue types identified in pancreas and prostate tumor samples.

Fig. 7

Box and whisker plots of (a) average scattered irradiance and (b) scatter power distributions, for all significant tissue types identified in pancreas tumor samples.

Fig. 8

Box and whisker plots of (a) average scattered irradiance and (b) scatter power distributions, for all significant tissue types identified in prostate tumor samples. No regions corresponding to early and intermediate fibrosis were identified in the prostate tumor samples.

Table 1

Histopathology-based classification of tissue subregions.