2.1.

Tissue Phantom Geometry and Optical Properties

Two-layer phantoms are used to mimic the structure of stratified cervical epithelium. The thickness of the epithelial (top) layer is set at $400\mathrm{\mu }\mathrm{m}$ to approximate the average thickness of human cervical epithelia.¹⁹ The stromal layer has a total depth of $25\mathrm{mm}$ to represent an optically semi-infinite tissue layer underneath the epithelium. The diameter of the cylindrical tissue phantoms is $22\mathrm{mm}$ , which is wide enough to encompass all optical paths collectible by our particular fiber geometry. A fine circular microscope cover glass (approximately $100\mathrm{\mu }\mathrm{m}$ thick) separates the two phantom layers, and all contact joints are sealed with an optical-grade adhesive. The influence of the glass separator on photon propagation in the tissue phantoms is empirically determined to be negligible. This conclusion is based on the comparison between the reflectance spectra of a homogenous phantom (with only one layer) and a two-layer phantom that consists of the same properties as those of the single-layer phantom.²⁰ The reflectance spectra of the two phantoms are almost indistinguishable for all the fiber geometries under consideration in this study. It should be noted that the glass separator may enhance the effective thickness of the top phantom layer. However, since the thickness of the cover glass is reasonably small and optically transparent, we believe that its inclusion is a reasonable and necessary compromise in the development of this two-layer phantom structure.

The optical properties of human cervical epithelium are based on the values reported by Collier, ²¹ Chang, ²² and Drezek ³ that were previously adopted by Zhu, Liu, and Ramanujam,¹⁰ Wang, ¹⁷ Skala, ¹⁸ and Palmer and Ramanujam²³ in their respective studies on epithelial tissue. A wide range of tissue optical properties is implemented in our phantoms to ensure the comprehensiveness and validity of our study. The values for the reduced-scattering $\left({\mu }_s^{\prime }\right)$ and absorption $\left({\mu }_a\right)$ coefficients of the normal and abnormal epithelial phantoms can be found in Figs. 1a and 1b , respectively. The reduced-scattering coefficients and anisotropies $\left(g\right)$ of the phantoms are calculated with Mie theory, and the absorption coefficients are experimentally determined using a spectrophotometer.

Fig. 1

(a), (b), and (c) The optical properties of the normal and abnormal epithelial phantoms that mark the boundary of the absorption and reduced-scattering coefficients implemented in the tissue phantoms. (c) depicts the reduced-scattering coefficients and anisotropies of the 0.51- and $0.92\text{-}\mathrm{\mu }\mathrm{m}$ polystyrene microspheres used in the basal layer.

2.2.

Fiber Geometry

The configuration of the source and collection fibers is controlled with a mechanical stage that independently adjusts the fiber separation (SDSD) and fiber obliquity with respect to the phantoms. The SDSD is measured as the distance between the opposing edges of the source and collection fibers. Throughout this study, we maintain the source fiber at the orthogonal orientation and only investigate the angles of the collection fiber in relation to the target surface. Fiber separation and obliquity are tuned with micrometers, of which the operational uncertainties are smaller than $1\mathrm{\mu }\mathrm{m}$ and $0.5\mathrm{deg}$ in magnitude, respectively. A circular cover glass is placed on the top of the phantoms to provide surface consistency, and the distal facets of the fiber optics are placed flush to the cover glass. Index matching between the fibers and cover glass is achieved with certified low-fluorescence immersion oil with a refractive index of 1.57 (Zeiss, Münich, Germany). The source and collection fibers, of which the core diameter and nominal numerical aperture are $200\mathrm{\mu }\mathrm{m}$ and 0.22, respectively, are connected to a broadband light source (Ocean Optics, Dunedin, Florida HR2000) and spectrophotometer (Ocean Optics, Dunedin, Florida QE64000), respectively. All reflectance measurements are normalized to the reference reflectance of a 99% reflectance standard (LabSphere, North Sutton, New Hampshire) when measured with the orthogonal $\left(0\mathrm{deg}\right)$ fiber. All reported reflectance spectra are sampled at three random locations on the phantoms and then averaged; the standard deviations with respect to the averages are within 5 percentage points for all measurements. For each measurement, five samples are automatically recorded and averaged with an integration time of $200\mathrm{ms}$ . To maintain phantom consistency, all phantoms are sufficiently agitated before each measurement.

2.3.

Phantom Properties and Rationale

This study investigates the sensitivities of fiber probes to: 1. changes in epithelial scattering properties, 2. changes in stromal scattering properties, and 3. changes in tissue scattering and absorption coefficients of both phantom layers that simulate dysplastic progression in cervical epithelial tissue. Mono-dispersed polystyrene microspheres (Bangs and Polysciences Laboratories) and 10% intralipid emulsions (Sigma-Aldrich) are the sources of scattering for the phantoms. For each of the investigations reported, specific combinations of scattering particles are chosen to stimulate the progression of changes in scattering properties of either epithelial or stromal phantom layers with respect to developing epithelial dysplasia.

3.1.

Changes in Epithelial Layer Scattering

This section investigates the ability of fiber probes to detect morphological variations of the scattering particles embedded in the phantoms’ top layer. As previously noted, epithelial dysplasia is linked to the growth of epithelial nuclei. The enlarged nuclei in the epithelial layer, in turn, increase the local scattering coefficients of the affected region. More importantly, such changes in nuclear morphology may then result in reflectance spectra that may potentially yield diagnostically significant information about tissue pathology. Thus, to the extent that a given fiber geometry is able to respond to slight modifications in tissue pathology, the sensitivity and specificity of optical diagnosis of tissue pathology may, in fact, be improved.

Figures 2, 3, 4, 5, 6 incorporate the reflectance spectra measured with various fiber angles and separations when only the epithelial (top) phantom layer is the subject of investigation. The properties of the phantoms correspond to the descriptions previously included in Sec. 2.3.1. Figure 2 shows the reflectance spectra of the four phantoms using the orthogonal fiber geometry (0-deg collection angle) at the SDSD of $500\mathrm{\mu }\mathrm{m}$ . It can be seen that the reflectance levels of the spectra gradually diminish against the trend of increasing sizes of polystyrene microspheres embedded in the epithelial layer. However, this trend is no longer true when we reduce the fiber separation from 500 to $200\mathrm{\mu }\mathrm{m}$ . For this fiber configuration (data shown in Fig. 3), the $4.65\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantom, which previously yielded the strongest reflectance, as shown in Fig. 2, now displays more subdued reflectance (third in magnitude; see Fig. 3) in relation to the other phantoms in the group. On the other hand, the 5.63- and $7.32\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantoms, which previously ranked as second and third in reflectance magnitude, respectively, now produce the two highest levels of reflectance with this orthogonal fiber geometry at a reduced SDSD. The significance of this transition is collectively explained in Sec. 4.1.

Fig. 2

Reflectance spectra detected by the 0-deg collection fibers at $\mathrm{SDSD}=500\mathrm{\mu }\mathrm{m}$ when the dimension of the particles in the epithelial layer increases from 4.65 to $9.0\mathrm{\mu }\mathrm{m}$ . The inset depicts the normalized spectra to demonstrate the congruence in spectral structures among the phantoms.

Fig. 3

Reflectance spectra detected by the 0-deg collection fibers at $\mathrm{SDSD}=200\mathrm{\mu }\mathrm{m}$ when the dimension of the particles in the epithelial layer increases from 4.65 to $9.0\mathrm{\mu }\mathrm{m}$ . The inset depicts the normalized spectra to demonstrate the congruence in spectral structures among the phantoms.

Fig. 4

Reflectance spectra detected by the 20-deg collection fibers at $\mathrm{SDSD}=200\mathrm{\mu }\mathrm{m}$ when the dimension of the particles in the epithelial layer increases from 4.65 to $9.0\mathrm{\mu }\mathrm{m}$ . The inset depicts the normalized spectra to demonstrate the spectral structures of the phantoms.

Fig. 5

Reflectance spectra detected by the 30-deg collection fibers at $\mathrm{SDSD}=200\mathrm{\mu }\mathrm{m}$ when the dimension of the particles in the epithelial layer increases from 4.65 to $9.0\mathrm{\mu }\mathrm{m}$ . The inset depicts the normalized spectra to demonstrate the spectral structures among the phantoms.

Fig. 6

(a) to (d) Reflectance spectra, equalized to 1 at $950\mathrm{nm}$ , of the two-layer phantoms, with microspheres of sizes 4.65, 5.63, 7.32, and $9.0\mathrm{\mu }\mathrm{m}$ , detected with increasing collection angles, demonstrating spectral differences between orthogonal and oblique fiber geometries. Phantoms with stronger upper-layer scattering, or larger particles, in this case, show faster structural convergences with increasing fiber obliquity. The fiber separation, measured edge to edge, is $200\mathrm{\mu }\mathrm{m}$ for all data shown.

To increase the probing sensitivity to the superficial layer, we incrementally increase the collection fiber angle to 20, 30, and $40\mathrm{deg}$ . Having the collection angle at $20\mathrm{deg}$ , with an SDSD of $200\mathrm{\mu }\mathrm{m}$ , the spectral characteristics of the polystyrene microspheres embedded in the top layers become immediately apparent, as illustrated in Fig. 4. The oscillations, specific to the sizes of the microspheres, are also apparent. Despite the clear signs of top-layer scattering in the spectra, the influence of the basal-layer reflectance is still evident. More specifically, the depression and peak at about the 575- and $675\text{-}\mathrm{nm}$ wavelengths, respectively, are still apparent in the spectra. The same characteristics in the spectral structures are also seen in Fig. 3, where the basal reflectance dominates. It should be noted that both the 4.65- and $5.63\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantoms show strong basal reflectance, which results from their low optical turbidity in the epithelial layer that allows an increased abundance of photons to reach the stromal layer.

One interesting observation of Fig. 4 is the reduced reflectance ripples for the $7.32\text{-}\mathrm{\mu }\mathrm{m}$ microspheres. The subdued oscillatory pattern of this particular phantom may be due to the cross-linker on the microspheres. This is true because, based on the results obtained from the scanning electron microscope and dynamic scattering spectroscopy, neither the morphological uniformity nor size dispersion (standard $\mathrm{deviation}=0.54\mathrm{\mu }\mathrm{m}$ , or equivalently 7.2% with respect to the mean diameter) of the $7.32\text{-}\mathrm{\mu }\mathrm{m}$ microspheres is affected by the added surface cross-linkers. The size dispersions for the other polystyrene microspheres, specifically for sizes 4.65, 5.63, and $9.0\mathrm{\mu }\mathrm{m}$ , are 0.63, 0.69, and $0.80\mathrm{\mu }\mathrm{m}$ , respectively (equivalently, 13, 12, and 8.9% with respect to their mean diameters). Based on these numbers, the cross-linked $7.32\text{-}\mathrm{\mu }\mathrm{m}$ microspheres do not suffer increased size variability in comparison to the other particles used in this study. It should also be noted that the $7.32\text{-}\mathrm{\mu }\mathrm{m}$ microspheres have the smallest size dispersion of the group. Since all other microspheres with broader size distributions can produce perfectly visible scattering oscillations (Fig. 4), there is no logical reason to expect otherwise from the $7.32\text{-}\mathrm{\mu }\mathrm{m}$ microspheres if the size dispersion were the acting factor here. More importantly, in a separate measurement of cross-linked $4\text{-}\mathrm{\mu }\mathrm{m}$ microspheres, we have also observed similarly diminished scattering oscillations in comparison to plain $4\text{-}\mathrm{\mu }\mathrm{m}$ microspheres. Therefore, we believe that the surface chemistry of the cross-linked particles may cause the diminished oscillations in the reflectance spectra. In spite of the reduced ripples for the cross-linked particles, the use of the $7.32\text{-}\mathrm{\mu }\mathrm{m}$ microspheres is justifiable to best meet the objectives of our study.

To further diminish the influence of basal-layer reflectance, we increase the collection angle to $30\mathrm{deg}$ , at which angle we simultaneously observe significantly enhanced scattering characteristics of the polystyrene microspheres in the top layer. That is, as the collection angle increases to $30\mathrm{deg}$ , the sequence of reflectance magnitude is at once aligned with the scattering intensities of the superficial layer, as shown in Fig. 5. In addition, the previously observed spectral depressions and peaks, at their respective wavelengths of 575 and $675\mathrm{nm}$ , are strongly suppressed. Significantly, the two prior results demonstrate that the $30\text{-}\mathrm{deg}$ collection fiber has sufficiently overcome the influence from the stromal layer and is predominantly sensitive to the scattering environments at superficial depths.

Figures 6a to 6d demonstrate the spectral progression of the phantoms measured with different fiber angles in the form of normalized reflectance spectra. The characteristics of the stromal-layer scattering particles, such as the valleys and peaks seen in Figs. 3 and 4, gradually disappear as the degree of fiber obliquity increases. For both the 4.65- and $5.63\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantoms, the low scattering coefficients of the epithelial layers, and resulting strong stromal layer reflectance, combine to permit a strong presence of stromal reflectance in the spectra so that the stromal signal only begins to phase out from the spectra when the collection angle reaches $35\mathrm{deg}$ . In contrast, the 7.32- and $9.0\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantoms yield virtually identical spectra when the collection angle only increases to $20\mathrm{deg}$ . Since the spectral structures of these particular phantoms are virtually invariant for fiber angles beyond $20\mathrm{deg}$ , we conclude that a distinct layer separation can be realized for the 7.32- and $9.0\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantoms at a fiber angle of $20\mathrm{deg}$ or more.

3.2.

Changes in Stromal Layer Scattering

Figure 7 shows the reflectance spectra sampled with the 0-deg fiber at the SDSD of $500\mathrm{\mu }\mathrm{m}$ . This configuration yields probing depths into the second phantom layer sufficient to resolve the difference in stromal scattering properties among the phantoms. A considerable disparity in the reflectance levels, as a result of the difference in intralipid concentration, is seen. In addition, the normalized reflectance spectra are also noticeably disparate between the two intralipid phantoms (see the inset of Fig. 7). More specifically, the 0.6% intralipid emulsion, because of its larger scattering coefficients, has a nearly flat reflectance spectrum, while the 0.3% intralipid emulsion yields relatively stronger reflectance in the blue region of the spectrum. However, we observe a change in the spectral structure of the intralipid reflectance when the fiber separation is reduced to $200\mathrm{\mu }\mathrm{m}$ (see Fig. 8 ). First, the reflectance spectrum of the 0.6% intralipid emulsion no longer appears to be flat. In addition, the intensity levels of the reflectance yielded by the intralipid phantoms are in better agreement with the 1:2 ratio of the emulsions (0.3 to 0.6%). Furthermore, the normalized spectra of the two phantoms, as shown in the inset of Fig. 8, are more structurally congruent, in that the normalized spectra virtually overlap each other for wavelengths longer than $550\mathrm{nm}$ . Consequently, the remnant disparity is only observable at the shorter end of the spectrum where the probing depths of the fibers are still too deep to effectively account for all superficially scattered rays in the 0.6% intralipid phantom. The significance of these observations relating fiber geometry and intralipid concentration in the basal layer of the phantoms is extensively analyzed in Sec. 4.2.

Fig. 7

Reflectance spectra detected by the 0-deg collection fibers at $\mathrm{SDSD}=500\mathrm{\mu }\mathrm{m}$ for various stromal scattering particles. The inset depicts the normalized spectra to demonstrate spectral structures of the phantoms that reflect the characteristics of the scattering coefficients and anisotropies of the basal layer.

Fig. 8

Reflectance spectra detected by the 0-deg collection fibers at $\mathrm{SDSD}=200\mathrm{\mu }\mathrm{m}$ for various stromal scattering particles. The inset depicts the normalized spectra to demonstrate spectral structures of the phantoms that reflect the characteristics of the scattering coefficients and anisotropies of the basal layer.

Returning to Figs. 7 and 8, we investigate another set of phantoms that incorporates 0.51- and $0.92\text{-}\mathrm{\mu }\mathrm{m}$ polystyrene microspheres as the source of scattering from the stromal layers of the tissue phantoms. Beginning with the 0-deg fibers separated with the SDSD of $500\mathrm{\mu }\mathrm{m}$ (data shown in Fig. 7), a fiber configuration that is expected to probing extensively into the phantoms, we observe that the reflectance difference between the 0.51- and $0.92\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantoms is generally only discernable at wavelengths greater than $750\mathrm{nm}$ . More interestingly, the two spectra intersect at approximately $550\mathrm{nm}$ , at which wavelength junction the reduced-scattering coefficients of the two phantoms are nearly identical [Fig. 1c]. On the other hand, when the fiber separation is reduced to $200\mathrm{\mu }\mathrm{m}$ , the $0.51\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantom renders stronger reflectance than the one that consists of the $0.92\text{-}\mathrm{\mu }\mathrm{m}$ microspheres in its stromal layer at the shorter fiber separation.

Similar to the results shown in Sec. 3.1, as we modify the collection angle to 20, 30, and $40\mathrm{deg}$ , the characteristics of the top phantom layers become more apparent (Fig. 9 ). Also, the oscillatory spectral patterns become more prominent as the fiber angle increases, and the spectral congruence among the curves is inversely dependent on the strength of basal scattering. The 0.3% intralipid yields the fastest convergence in terms of reflectance intensity as the fiber obliquity increases, whereas the 0.6% intralipid is the slowest. For demonstrative purposes, Fig. 10 shows that good conformities in both reflectance levels and spectral characteristics, specific to the epithelial phantom layers, are achievable with the 40-deg collection angle, regardless of the different scattering compositions used in the stromal phantom layers.

Fig. 9

(a) to (d) Reflectance spectra, equalized to 1 at $950\mathrm{nm}$ , of the two-layer phantoms, with 0.51- and $0.92\text{-}\mathrm{\mu }\mathrm{m}$ polystyrene microspheres and 0.3 and 0.6% intralipid emulsions in the basal layer, detected with increasing collection angles, demonstrating spectral differences between orthogonal and oblique fiber geometry. Phantoms with greater basal-layer scattering show faster structural convergences with increasing fiber obliquity. The fiber separation, measured edge to edge, is $200\mathrm{\mu }\mathrm{m}$ for all data shown.

Fig. 10

Reflectance spectra detected by the 40-deg collection fibers at $\mathrm{SDSD}=200\mathrm{\mu }\mathrm{m}$ when different stromal scattering particles are used. The inset depicts the normalized spectra to demonstrate the congruence in spectral structures among the phantoms. Similar spectra are seen regardless of the composition of the basal layers.

3.3.

Changes in Tissue Scattering and Absorption Simulating Epithelial Dysplasia

As extensively detailed in Sec. 2.3.3, four tissue phantoms, which include hemoglobin, are created to investigate the performance of depth-selective fiber geometry. Six fiber configurations, to include (0-deg collection at 200- and $500\text{-}\mathrm{\mu }\mathrm{m}$ SDSDs); (20-deg collection at 200- and $500\text{-}\mathrm{\mu }\mathrm{m}$ SDSDs); and {40-deg collection at 200-, and $500\text{-}\mathrm{\mu }\mathrm{m}$ SDSDs}; are used for the reflectance measurements.

Figure 11 shows the reflectance collected from these phantoms using the 0-deg collection fiber with $500\text{-}\mathrm{\mu }\mathrm{m}$ separation. Using the orthogonal geometry and a large fiber separation of $500\mathrm{\mu }\mathrm{m}$ , a configuration expected to produce extended probing depths, phantoms that have different stromal layers are well differentiated in the reflectance spectra, while the spectra of phantoms with common stromal layers, such as the $N$ - $N$ , $A$ - $N$ , and $A$ - $A$ , $N$ - $A$ pairs, bear strong resemblances in both reflectance level and spectral structure. For this particular fiber geometry, this result suggests that differences in the epithelial layers do not affect the overall reflectance measurements. However, we have demonstrated that more superficially sensitive probing can be achieved by reducing the fiber separation distance to $200\mathrm{\mu }\mathrm{m}$ . Figure 12 and its inset show that a reduction in probing depths does result in stronger spectral sensitivities to the epithelial phantom layers. This is supported by comprehensive evidence that: 1. the reflectance valley induced by hemoglobin’s Soret absorption (at $415\mathrm{nm}$ ), in relation to the whole spectrum, is reduced—a result that signifies a relative decrease in hemoglobin absorption; 2. the proportion of reflectance in the blue region of the spectrum has increased—an indication of increased detection of superficially penetrating photons; and 3. the spectra, which previously correlated solely to the stromal compositions of thephantoms, now display significant distinctions, even for phantoms with common stromal layers.

Fig. 11

Reflectance spectra detected by the 0-deg collection fibers at $\mathrm{SDSD}=500\mathrm{\mu }\mathrm{m}$ with respect to different combinations of phantom layers. For this deeply probing fiber geometry, phantoms sharing common basal layers yield similar reflectance spectra in both reflectance levels and spectral structures. The difference in the upper layer is not clearly resolved by this fiber geometry.

Fig. 12

Reflectance spectra detected by the 0-deg collection fibers at $\mathrm{SDSD}=200\mathrm{\mu }\mathrm{m}$ with respect to different combinations of phantom layers. For this deeply probing fiber geometry, phantoms sharing common basal layers yield similar spectral structures. The reflectance levels are strongly influenced by the scattering properties of the upper phantom layers; however, the difference in the upper layer is not reflected in the normalized spectra.

The spatial balance of the collected reflectance shifts toward the superficial layer whenever increasingly oblique collection angles are used. The 20-deg $500\text{-}\mathrm{\mu }\mathrm{m}$ fibers have reflectance spectra similar to those of the 0-deg $200\text{-}\mathrm{\mu }\mathrm{m}$ fibers. Since the general observations made for the 20-deg $500\text{-}\mathrm{\mu }\mathrm{m}$ fibers are consistent with those for the 0-deg $200\text{-}\mathrm{\mu }\mathrm{m}$ fibers, the graphical data are not repeated here. A more interesting result comes from the 20-deg $200\text{-}\mathrm{\mu }\mathrm{m}$ fiber pairs, of which the reflectance spectra simultaneously exhibit dual characteristics of both tissue layers. In all the spectra shown in Fig. 13 , the fine oscillatory spectral patterns, specific to the types of polystyrene microspheres embedded in the epithelial phantom layers, are visually distinguishable, while the coarse structures of the spectra are commonly shared by phantoms with the same stromal layers.

Fig. 13

Reflectance spectra detected by the 20-deg collection fibers at $\mathrm{SDSD}=200\mathrm{\mu }\mathrm{m}$ with respect to different combinations of phantom layers. Fine oscillatory patterns specific to the microspheres embedded in the upper layer can be seen in their respective phantoms. The coarse structures of the overall spectra still largely depend on the composition of the basal layer. The grouping of the spectra has been swapped for this superficially probing fiber geometry.

It is apparent that layer separation in the reflectance spectra correlates to the degree of fiber obliquity; thus, the use of a more obliquely oriented collection fiber will, correspondingly, achieve a more definitive separation of the layers in the reflectance spectra. The results shown in Fig. 14 demonstrate that the 40-deg. fiber with $200\text{-}\mathrm{\mu }\mathrm{m}$ SDSD is sufficiently capable of resolving the morphology of the epithelial scattering particles and eliminating the hemoglobin interferences from the basal layers. The 40-deg $500\text{-}\mathrm{\mu }\mathrm{m}$ fibers share similar characteristics with the 20-deg $200\text{-}\mathrm{\mu }\mathrm{m}$ fibers, and thus are not repeated here. In Fig. 14, the oscillatory patterns, specific to their respective polystyrene microspheres embedded in the epithelial layers, are clearly resolved. In addition, signs of hemoglobin absorption, especially at the Soret $\left(415\mathrm{nm}\right)$ and Q-band wavelengths (540 and $580\mathrm{nm}$ ), are strongly suppressed.

Fig. 14

Reflectance spectra detected by the 40-deg collection fibers at $\mathrm{SDSD}=200\mathrm{\mu }\mathrm{m}$ with respect to different combinations of phantom layers. Definitive oscillatory patterns specific to the microspheres embedded in the upper layer can be seen in their respective phantoms despite the phantom composition in the basal layers.

4.1.

Depth-Resolved Reflectance: Sensitivities to Morphological Variations in the Superficial Layer

From the results presented in Sec. 3.1, particularly for the orthogonal fiber geometry with $500\text{-}\mathrm{\mu }\mathrm{m}$ SDSD, we observed an opposing trend in the intensity level of the reflectance spectra with respect to the scattering coefficients of the epithelial phantom layer. The diminishing reflectance levels, together with the opposing phenomenon of increasing epithelial scattering due to enlarged scattering particles, lead to the conclusion that this particular fiber geometry might be predominantly sensitive to the stromal phantom layer, rather than the epithelial layer. This is true because, with greater epithelial scattering in the phantoms, it is likely that more photons are scattered and reflected at shallow depths and thus not detected by the collection fiber situated far from the illumination point. By the same argument, the rise in superficial scattering should reduce the number of photons that reach the stromal layer, from which this particular fiber configuration gathers the majority of its reflectance signal. As a consequence, the diffuse reflectance originating from the stromal layer decreases, which results in the diminishing reflectance levels we observe. The increase in the number of early-returning photons, caused by the rise in epithelial scattering, is not seen here because the extended SDSD precludes the collection of photons that are reemitted from shallow tissue depths. There is more evidence to indicate that this particular fiber geometry cannot differentiate the enlarging scattering particles in the top layer. Specifically, and importantly, among the four phantoms, the reflectance levels only differ within 3 percentage points, which indicates that it is difficult to differentiate among the four phantoms based on the intensity levels of the reflectance spectra. Furthermore, the spectral characteristics, as shown in the inset of Fig. 2 in the form of normalized reflectance spectra, are nearly uniform. This indicates that the reflectance spectra share similar properties and are of the same origin (the stromal phantom layer). The results shown in Fig. 2 give little indication of the enlarging scattering particles in the epithelial layer, while all other evidence suggests a uniform origin of the reflectance spectra. This affirms our expectation of a deeply probing fiber geometry characterized by minimal sensitivity to the epithelial layer.

Thus, to execute a more superficially sensitive reflectance measurement using the orthogonal fiber geometry, the SDSD is reduced to $200\mathrm{\mu }\mathrm{m}$ . For this fiber configuration, we find some evidence, as shown in Fig. 3, which suggests increased probing sensitivity to the upper layer. Specifically, we observed a shift in the levels of reflectance from the phantoms detectable by the fibers when the SDSD is reduced from 500 to $200\mathrm{\mu }\mathrm{m}$ . From the data reported in Sec. 3.1, the $4.65\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantom, which yields the strongest reflectance when the SDSD is $500\mathrm{\mu }\mathrm{m}$ , displays more subdued reflectance (third in magnitude relative to the other phantoms) when the SDSD is reduced to $200\mathrm{\mu }\mathrm{m}$ . The significant decrease in reflectance of the $4.65\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantom is expected for this particular fiber geometry. This is because the $4.65\text{-}\mathrm{\mu }\mathrm{m}$ microspheres offer the lowest epithelial scattering coefficients of the phantom group; consequently, the number of reflected photons from the epithelial layer for the same phantom is also expected to be lowest. In addition, as the sensitivity of the fiber probe shifts toward the epithelial layer, the determination of the overall reflectance no longer solely depends on the reflectance from the stromal phantom layer. Consequently, with the combined effect of minimal epithelial photons and lessened sensitivity to the stromal layer, the strength of reflectance of the $4.65\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantom decreases significantly in comparison to the other phantoms. On the other hand, the phantom with the $9.0\text{-}\mathrm{\mu }\mathrm{m}$ microspheres in its top layer represents the opposite extreme where the increase in superficial-layer scattering, which strongly attenuates the number of photons penetrating to the stromal layer, does not sufficiently compensate for the reduction in basal reflectance. This is because a large proportion of the fibers’ sensitivity still resides in the stromal layer for the orthogonal fiber geometry. Not surprisingly, the intermediate levels of epithelial scattering of the 5.63- and $7.32\text{-}\mathrm{\mu }\mathrm{m}$ microspheres should and did result in the maximal detection of the overall reflectance. This conclusion may be explained by the presumption that these microspheres would necessarily augment reflectance originating from the epithelial layer before excessively reducing light reflected from the stroma. This indicates that the dual reflectance contribution from both the top and bottom layers of the phantoms can be attributed to the orthogonal fiber configuration when the SDSD is small. Although the intensity of the reflectance spectra is affected by the scattering coefficients of the epithelial phantom layer, as discussed before, the spectral structures of the normalized spectra, as shown in the inset of Fig. 3, remain uniform. The uniformity of spectral structure implicates a singular dominant source of the reflectance for all four phantoms. Since the only common denominator among the phantoms is the invariant stromal layer, we conclude that the orthogonal fiber geometry predominantly interrogates the stromal layer and does not sufficiently resolve the enlarging scattering particles at superficial depths.

Using the 20-deg fiber (Fig. 4), the $7.32\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantom exhibits the highest reflectance, even though the $9.0\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantom has the highest epithelial scattering coefficients among the four phantoms. This phenomenon may also be attributed to the dual-layer probing of the fiber. While the $9.0\text{-}\mathrm{\mu }\mathrm{m}$ microspheres supply the strongest reflectance from superficial depths, the correspondingly converse effect of reducing basal reflectance is also significant. As the 20-deg fiber still collects reflectance from both phantom layers, the reduction in stromal-layer reflectance still negatively affects the overall reflectance level of the $9.0\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantom. However, as the fiber angle continues to increase, the resultant reflectance spectra become more characteristic of the scattering microspheres embedded in the epithelial layer. More specifically, at the collection angle of $30\mathrm{deg}$ , the sequence of reflectance magnitude agrees with the increasing trend of epithelial scattering coefficients in the top phantom layers. Furthermore, the stroma-specific spectral depressions and peaks, particularly observable at the wavelengths of 575 and $675\mathrm{nm}$ , are strongly suppressed in the data shown in Fig. 5. The combined results indicate that the reflectance signal from superficial depths can be effectively enhanced, even in the presence of a dominant reflectance source, such as the stromal layer implemented in our phantoms.

4.2.

Depth-Resolved Reflectance: Sensitivities to Changes in the Morphology of the Scattering Elements in the Basal Layer

From the results presented in Sec. 3.2, particularly for those involving the intralipid phantoms, we observed that the spectral structures and intensity levels of the reflectance spectra vary significantly with respect to fiber geometry. Beginning with the orthogonal fiber with SDSD of $500\mathrm{\mu }\mathrm{m}$ , the high scattering of the 0.6% intralipid emulsion produces reflectance that is evenly distributed across the spectrum in spite of the fact that photons are more intensely scattered at shorter wavelengths. This preferential enhancement of the long wavelengths is most logically attributed to the particular fiber geometry that is in use. When the SDSD is sufficiently extensive, the rise in the number of photons reflected near the source fiber is less detectable to the collection fiber, which is particularly sensitive to deeper depths of the phantom. Therefore, photons near the blue end of the spectrum, although they are in fact more intensely scattered and reflected nearer the phantom surface than those near the NIR region, do not proportionally contribute an equal gain in reflectance as do their counterparts in the NIR. Consequently, the spectral structure flattens when the intralipid concentration increases from the 0.3 to 0.6% emulsion. The measured reflectance spectra further support this analysis when the fiber separation decreases to $200\mathrm{\mu }\mathrm{m}$ (data shown in Fig. 8), since the fiber geometry becomes more sensitive to superficially scattered photons at shallower depths. More specifically, the intensity levels of the reflectance yielded by the intralipid phantoms are in better agreement with the 1:2 ratio of the emulsions (0.3 to 0.6%). This indicates that the effect of omitting superficially scattered photons of the blue wavelengths, as a result of the higher phantom scattering coefficients, is not as pronounced as in the previous instance where the reflectance levels of the intralipid phantoms are very similar (differing by less than a percentage point) at the 400-nm wavelength. The greater sensitivity to shorter spectral wavelengths suggests that reduction in SDSD does, indeed, moderately lift the probing focus of the fiber probe from the deep stromal layer toward the phantom surface.

The intralipid phantoms allow us to test the capability of fiber probes to resolve concentration differences, while the polystyrene phantoms, namely those with the 0.51- and $0.92\text{-}\mathrm{\mu }\mathrm{m}$ microspheres in the basal layers, are more useful for examining reduced-scattering coefficients and anisotropies. The anisotropy is of great interest to us because it is strongly dependent on the morphology of the scattering elements. For example, the anisotropy of spherical particles is correlated to the sizes of the particles: particles greater in size typically have anisotropies in favor of the forward direction (thus, of values nearer to 1), while smaller particles are generally less forward scattering. Since these two phantom types have similar reduced-scattering coefficients $\left({\mu }_s^{\prime }\right)$ , it should be noted that the $0.92\text{-}\mathrm{\mu }\mathrm{m}$ polystyrene phantom has greater scattering coefficients in the natural form $\left({\mu }_s\right)$ than the $0.51\text{-}\mathrm{\mu }\mathrm{m}$ phantom, because the anisotropy factors $\left(g\right)$ of the former are also greater. Taking these findings into consideration, we believe that the reduced-scattering coefficients are a more descriptive metric to describe the scattering properties of the phantoms when the fiber geometry is primarily sensitive to deeper depths. Under these circumstances, the mathematical models of diffuse reflectance and the use of reduced-scattering coefficients in the description of bulk reflectance are better observed. In contrast, when the fiber geometry is more sensitive to superficial scattering, anisotropic scattering is specifically amplified in the determination of reflectance spectra. Under this set of circumstances, large angular deflections, during the propagation of scattered photons, are relevant.¹³ In other words, the $0.51\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantom, by virtue of its smaller anisotropies, is expected to produce markedly stronger reflectance than the $0.92\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantom when the fibers’ probing depths are superficial. Conversely, both of the microsphere phantoms would generate reflectance of similar magnitude when the fibers’ probing depths are sufficiently deep.

The results presented in Sec. 3.2 confirm our hypothesis stated earlier. More specifically, the reflectance difference between the 0.51- and $0.92\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantoms is generally discernable only at wavelengths greater than $750\mathrm{nm}$ when the orthogonal fiber geometry with $500\text{-}\mathrm{\mu }\mathrm{m}$ SDSD is used. More interestingly, the two spectra intersect at approximately $550\mathrm{nm}$ , at which wavelength junction the reduced-scattering coefficients of the two phantoms are nearly identical [Fig. 1c]. Beyond $550\mathrm{nm}$ , the reduced scattering of the $0.92\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantom is slightly greater than that of the $0.51\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantom, and this trend is correspondingly manifested in the spectra. A logical explanation may be that the $0.92\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantom, with its forward-scattering anisotropies and higher values in both the native- and reduced-scattering coefficients, yields particularly strong basal reflectance at deeper depths that are preferentially detected by the 0-deg $500\text{-}\mathrm{\mu }\mathrm{m}$ SDSD fiber configuration. While these matching observations could be merely coincidental, information obtained from the data cited in Fig. 8 strengthens our belief that the prominence of the anisotropy in the determination of reflectance increases as the fiber geometry becomes more sensitive to superficially scattered photons. The data given in Fig. 8 show that the $0.51\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantom renders stronger reflectance than the one that consists of the $0.92\text{-}\mathrm{\mu }\mathrm{m}$ microspheres in the stromal layer at $200\text{-}\mathrm{\mu }\mathrm{m}$ SDSD. It results from the natural tendency of photons to be backscattered by smaller particles, and the anisotropic scattering effect may offset the greater scattering power of the larger particles and push the intersection of the two spectra from $550\mathrm{nm}$ , as previously shown in Fig. 7, to nearly $800\mathrm{nm}$ . At longer wavelengths, where the probing depths of the fibers are sufficiently deep, the reflectance detection remains preferential to the $0.92\text{-}\mathrm{\mu }\mathrm{m}$ microsphere phantom, of which the higher anisotropies direct more photons forward into the medium, resulting in higher reflectance from the stromal layer. Based on our experimental observations, we believe that the anisotropies of the scattering particles in the phantoms could have a strong influence over the spectral dependence of the reflectance, especially at superficial probing depths where the phenomenon of backscattering dominates.

4.3.

Depth-Resolved Reflectance: Tissue Phantoms with Relevant Tissue Scattering and Absorption

Based on the results shown in Sec. 3.3, we observed that the orthogonal fiber geometry with extensive SDSD is able to differentiate the phantom properties of the stromal layer, whereas the difference in the epithelial phantom layer does not affect the overall reflectance spectra. However, as the SDSD is reduced to $200\mathrm{\mu }\mathrm{m}$ , we begin to observe the epithelial influence in the overall spectra. This result agrees with the conclusion drawn in our previous discussion: that the SDSD indeed has significant effect on the spatial distribution of the reflectance measurements. Based on the results shown in Fig. 12, we see a relative decrease in the reflectance intensity of phantoms with abnormal epithelial layers when compared to their counterparts with normal epithelial layers. This suggests that the influence from the epithelial layers may require consideration in the overall interpretation of the reflectance spectra. Specifically, since the scattering coefficients of the abnormal epithelium are 2.5 times higher than those of the normal epithelium, the probability of photons being superficially reflected from depths outside the detection of the collection fiber is then also higher. The stronger epithelial scattering, which reduces the number of photons reaching the stromal layer, consequently diminishes the reflectance originating from greater depths. However, despite the difference in reflectance due to the variations in the epithelial layers, the more interesting observation is that the normalized spectra of the phantoms with common stromal layers remain congruent. This finding indicates the inadvisability of using flush fibers to isolate the signals from superficial tissue depths in as much as the stromal layers remain a dominant part of the overall reflectance.

As the collection angle continues to increase to $40\mathrm{deg}$ , we are able to achieve complete layer separation by demonstrating that the reflectance spectra sampled by the 40-deg fibers only correlate to the epithelial phantom layers. This finding, as well as the results presented in Secs. 3.1, 3.2, all affirm the viability of using the angled-fiber approach to correctly characterize tissue with respect to particular layers of interest.

4.4.

Comparisons to Published Work on Spatially Resolved Spectroscopy

In recent years, the importance of spatially resolved spectroscopy has gained much interest. From Monte-Carlo-based computational analyses to clinical measurements, many different approaches have been proposed to achieve desired depth selection in stratified tissue structures. More specifically, with respect to fiber geometries, the commonly adopted approaches are fiber separation (SDSD) and angles. Amelink, ^{34, 35} in their differential path-length spectroscopy study, demonstrated the feasibility of achieving analysis of particle size in superficial layers of turbid media by sampling diffuse reflectance at different fiber separations, followed by a postmeasurement algorithm to compute the superficially scattered reflectance signals. Other studies on the approach of SDSD have also demonstrated that large fiber separations consistently produce extensively diffused reflectance and deeper probing depths.^{17, 36} More specifically, Arifler ³⁶ demonstrated an increasingly significant influence of hemoglobin absorption to the overall reflectance spectra as SDSD increases. Their result is consistent with the observations made in our study, particularly in Sec. 3.3.

In addition to SDSD, the approach of utilizing variable fiber angles to achieve spatially resolved spectroscopy has also been demonstrated. One advantage of the angular approach is its ability to further enhance the detection of superficially scattered light. As shown in this study, superficially scattered reflectance can be directly measured with angled fibers without further signal processing. Liu and Ramanujam,³⁷ and Skala ¹⁸ have also shown the possibility of detecting enhanced fluorescence signals from epithelial phantom layers as the illumination angle increases from 0 to $45\mathrm{deg}$ . Another interesting approach, while similar in principle to our angled fiber geometry, uses a ball lens-coupled fiber probe instead of rotating fibers for depth-resolved spectroscopy of epithelial tissue.^{15, 38} By adjusting the focal point of the ball lens, the illumination and collection angles are effectively manipulated. Thus, enhanced epithelial fluorescence and reflectance measurements are demonstrated in both computational prediction and phantom experimentation. Neiman ¹⁶ showed that the applicability of the angled fiber approach can be further improved when polarization gating is integrated in the fiber probe to provide more refined spatial resolution for superficially scattered signals.

While the geometric manipulation of fiber probes is indeed an effective means of achieving more refined spatial resolution in turbid media, the spatial selectivity of this approach ranges from 50 to a few hundred micrometers,^{17, 39} whereas the spatial resolutions of optical coherence gating and confocal imaging are approximately 10 to $15\mathrm{\mu }\mathrm{m}$ . However, this simple fiber geometry manipulation does provide sufficient spatial selectivity for spectroscopy where the spatial requirement is relatively minimal. Fiber geometries and their effectiveness for spatial selection are expected to significantly improve the efficacy of current spectroscopic diagnostics with limited additional implementation complexity and cost, which will result in accurate, fast, and portable optical diagnostics in the future.