2.1.

Overview

The study was reviewed and approved by the Institutional Review Boards at the University of Texas at Austin, the University of Texas M.D. Anderson Cancer Center, and Rice University. Fluorescence spectra were measured in vivo from clinically normal and abnormal oral sites using a fiber optic probe that preferentially collects fluorescent light generated in the epithelium and superficial stroma. Biopsies were obtained from measured sites, and high resolution fluorescence confocal microscopy was used to record autofluorescence patterns as a function of depth at each site. Using these site-specific fluorescence distributions as input to a Monte Carlo model, we predicted the tissue fluorescence spectrum for each site. Predicted fluorescence spectra for six oral sites with different pathological diagnoses and/or anatomical types were compared to the clinical spectra measured at each site.

2.2.

Depth-Sensitive Monte Carlo Modeling of Oral Mucosa

A fixed weight Monte Carlo program was used to simulate the collection of fluorescence spectra using a fiber optic probe designed to preferentially collect fluorescence generated in the epithelium and superficial stroma. The program was designed to account for generation and propagation of fluorescent light in multilayered oral tissue. Details regarding the Monte Carlo code and the depth-sensitive fiber optic probe geometry have been described previously.¹² Briefly, excitation photons were initially propagated through the tissue using the scattering coefficient $\left({\mu }_s\right)$ and absorption coefficient $\left({\mu }_a\right)$ at the excitation wavelength according to the layer they are traveling through. A photon has a probability of ${\mu }_a∕({\mu }_s+{\mu }_a)$ being absorbed at a scattering site. If a photon is absorbed by a fluorophore, an emission photon may be generated and is propagated according to the ${\mu }_a$ and ${\mu }_s$ for the emission wavelength in the layer in which it is traveling. The photons are propagated until they reach the surface or are reabsorbed. The number of photons that reach the surface and are detected by a detector are calculated. A single simulation output predicts the fluorescence emission at $350\text{-}\mathrm{nm}$ excitation, and consists of the detected fluorescence intensity (total number of detected fluorescence photons) at emission wavelengths of 400, 420, 440, 450, 460, 480, 500, 520, 540, and $560\mathrm{nm}$ . Each simulation was performed with ${10}^8$ excitation photons. For each detected photon, the depth at which fluorescence emission occurred was recorded. The fraction of detected photons originating in the epithelium and the fraction originating in the stroma were calculated.

2.3.

Clinical Measurements, Biopsy Collection, and High Resolution Tissue Imaging

Fluorescence emission spectra at $350\text{-}\mathrm{nm}$ excitation were measured from clinically normal and abnormal sites using a depth-sensitive fiber optic probe. Detailed descriptions of the clinical spectroscopic device, the depth-sensitive fiber optic probe, and procedures for patient recruitment, data acquisition, calibration, and processing have been reported in Ref. 5. The optical probe contains four depth channels, each consisting of a set of illumination and collection fibers arranged in a geometric configuration designed to preferentially collect light from a specific depth region in the tissue. The shallow channel, used to collect the clinical data presented here, uses a ball lens at the tip of the probe to redirect the angle of the illumination light such that its intersection with the collection region is localized to the tissue surface. This enables preferential collection of signal from the epithelium and superficial stroma.

Data from six sites in three patients were used to validate the Monte Carlo model. To explore variations in spectroscopic data with the presence of benign inflammation and neoplasia, data were selected from a patient with measurements made from four sites in the tongue with histologic diagnoses of normal, inflammation, dysplasia, and cancer. To explore variations with tissue anatomic sites, measurements were selected from two additional patients with measurements of normal palate and normal buccal mucosa. After spectroscopic measurement, each site was biopsied and the tissue sample was divided into two portions. One portion of each sample was sent for histologic examination. Transverse $200\text{-}\mu \mathrm{m}$ -thick tissue slices were immediately prepared from the second portion of each sample and imaged with a confocal fluorescence microscope while the tissue was still viable. Confocal images were obtained at ultraviolet (UV 350 and $364\mathrm{nm}$ ) excitation and at $488\text{-}\mathrm{nm}$ excitation to determine input parameters to the Monte Carlo model for that site. Detailed procedures for tissue preparation, confocal imaging, and quantitative analysis of autofluorescence patterns are described in Ref. 6.

2.4.

Tissue Geometry and Model Input Parameters

For normal, inflammatory, and dysplastic sites, tissue was modeled as a multilayered turbid medium with five sublayers: superficial epithelium, intermediate epithelium, basal epithelium, superficial stroma, and deep stroma.^{6, 12} Each sublayer was assumed to contain a single dominant fluorophore, characteristic of that sublayer and diagnosis. At each site, the UV confocal fluorescence images were analyzed to determine the thickness and average fluorescence intensity of each sublayer. These values were used as site-specific input to the Monte Carlo model.

Previous work has shown that the superficial epithelium includes cells that retain keratin, and the main fluorophore in this layer was assumed to be keratin.⁶ The primary fluorophore in the intermediate epithelium was assumed to be mitochondrial flavin adenine dinucleotide (FAD).⁶ The dominant fluorophore in the basal epithelium was assumed to be mitochondrial reduced nicotinamide adenine dinucleotide (NADH).⁶ For the normal and dysplastic sites, the primary fluorophores in both the superficial and deep stroma were assumed to be collagen crosslinks.¹³ In oral lesions with severe stromal inflammation, confocal microscopy studies demonstrate stromal fluorescence that appears to originate from inflammatory cells with strong cytoplasmic fluorescence at UV excitation. Thus for the inflammatory site considered in this study, the dominant stromal fluorophore was assumed to be mitochondrial NADH.

The site diagnosed as invasive cancer was modeled using a different tissue geometry, because the normal structure is lost with invasion. This site was modeled as a turbid medium with cellular and fibrous layers. The geometry was derived from visual examination of the confocal images obtained from the measured site. Five sublayers were defined: cellular region 1, with dominant keratin fluorescence and optical properties similar to noncancerous superficial epithelium; cellular region 2, with weak cytoplasmic fluorescence from FAD; cellular region 3, with bright cytoplasmic fluorescence from NADH; matrix region 1, with dominant NADH fluorescence associated with a significant cellular component consisting of cancer, inflammatory, and atypical cells; and matrix region 2, a fibrous region with collagen as the primary fluorophore and optical properties similar to those of noncancerous deep stroma.

The scattering coefficients of the epithelial and stromal layers were obtained from previous studies of oral, cervical, and bronchial tissue. The values of the scattering coefficients at $350\mathrm{nm}$ and absorption coefficients at $420\mathrm{nm}$ are summarized in Table 1 . The scattering coefficient of the intermediate and basal epithelium was derived from reflectance confocal images of normal, dysplastic, and cancerous nonkeratinized oral epithelium.¹⁴ The scattering coefficient of the superficial epithelium in all oral sites considered in this study was derived from reflectance confocal measurements of cervical keratinized epithelium.¹⁵ In normal oral tissue, the scattering coefficients for the superficial and deep stromal layers were assumed to be the same and were based on values from Ref. 16. To accurately model the drop in stromal scattering associated with oral dysplasia, the scattering coefficient in the superficial stroma in the dysplastic case was decreased by 20%.¹⁷ Scattering coefficients obtained from the literature were extrapolated to lower wavelengths using the inverse proportionality relationship of cellular scattering to wavelength.¹⁸

Table 1

Scattering and absorption coefficients for each oral tissue site and sublayer. Values were used as input to the Monte Carlo model.

The absorption coefficient for the epithelium was assumed to be constant for all sites considered in this study and was based on values derived in Ref. 16. The stromal absorption coefficient used in this model has contributions from three different absorbers: collagen, oxyhemoglobin, and deoxyhemoglobin. The collagen absorption coefficient was based on measurements from bloodless skin samples.¹⁹ The oxy- and deoxyhemoglobin absorption coefficients were calculated as described by Jacques.²⁰ An oxygen saturation level of 80% and a hemoglobin concentration of $150\mathrm{g}∕\mathrm{l}$ were used.

For the normal tongue site and the dysplastic tongue site, the volume fraction of blood in the stroma was determined from quantitative analysis of images from hematoxylin and eosin (HE) stained slides from that specific site. HE images of these sites were analyzed to estimate the fraction of the total stromal area occupied by blood. The estimated value included areas occupied by the lumen of blood vessels. The volume fraction of blood in the superficial stromal layer of the normal tongue site was estimated to be in the range 0.05 to 0.1%, while the value for deep stroma was in the range 0.2 to 0.3%. The superficial stromal layer underlying the epithelium in the dysplastic tongue site was estimated to have a volume fraction of blood in the range 0.10 to 0.20%, while the value for deep stroma was similar to that of the normal case. The values obtained for the normal tongue site were used for the other normal sites as well. For the cancer site and the inflammatory site, the volume fraction of blood in the stroma was estimated based on information from the literature. For the cancer site, the volume fraction of blood was estimated as four times higher than that of normal tissue, based on previous research indicating that the microvessel density of oral cancers is about three to four times higher than in normal oral tissue.²¹ For the inflammatory site, the volume fraction of blood was estimated to be similar to the blood volume fraction of normal tissue.

3.1.

Site-Specific Spectral Changes Caused by Neoplastic Progression in Oral Mucosa

Figure 1 shows confocal fluorescence images and measured clinical spectra obtained at UV excitation from four different sites on the tongue of a single patient. One of the four sites was clinically and histologically normal. The other three sites were oral lesions that were diagnosed histologically as inflammation, dysplasia, and cancer, respectively. The normal oral tissue is characterized by strong epithelial and stromal autofluorescence [Fig. 1a]. In contrast, the oral lesion diagnosed with severe inflammation displays a large drop in both epithelial and stromal fluorescence [Fig. 1b]. In addition, while stromal fluorescence in the normal tissue originates from collagen fibers, fluorescence from the inflammatory stroma originates predominantly from a large influx of inflammatory cells. The oral lesion diagnosed with dysplasia is characterized by increased epithelial thickness, brightly fluorescent cells throughout the epithelium, and a drop in fluorescence of the superficial stroma [Fig. 1c]. Confocal images of a malignant lesion diagnosed with moderately differentiated cancer [Figure 1d] illustrate the loss of layered architecture typical for cancers. In this example, the superficial $400\mu \mathrm{m}$ of the tissue are occupied by cancer cells; underneath is a region of matrix infiltrated by cancer and inflammatory cells.

Fig. 1

(a) to (d) Fluorescence confocal images and (e) measured fluorescence spectra at UV excitation, showing neoplastic progression in the tongue. (a) Normal tongue; (b) normal tongue with severe stromal inflammation; (c) focal mild dysplasia; and (d) moderately differentiated cancer. In (a), (b), and (c), the white lines represent the basement membrane and the yellow lines represent the tissue surface. Scale bars represent $200\mu \mathrm{m}$ . (e) Fluorescence spectra at $350\text{-}\mathrm{nm}$ excitation, measured from the same oral sites shown in the confocal images. Images (b) and (c) are reproduced from Ref. 6 with permission.

Clinically measured emission spectra at $350\text{-}\mathrm{nm}$ excitation from the four oral sites are displayed in Fig. 1e. The data illustrate the characteristic drop in overall intensity associated with neoplastic progression in oral mucosa. Spectra from the dysplastic and cancerous sites also have lower relative emission intensity in the $400\text{-}\text{to}430\text{-}\mathrm{nm}$ wavelength region, giving them a different spectral shape compared to the normal case. The spectrum from the inflammatory site displays both a drop in fluorescence intensity and a shift to a longer peak emission wavelength compared to the normal spectrum.

Table 2 summarizes the site-specific tissue properties that were derived from the confocal fluorescence images of each site. At a given site, each of the five tissue sublayers was assigned a thickness, a dominant fluorophore, and a relative fluorescence intensity based on quantitative analysis of the confocal fluorescence images.

Depth-resolved spectra for all tissue sites were simulated with the Monte-Carlo-based model at $350\text{-}\mathrm{nm}$ excitation using the input values shown in Tables 1, 2. Figure 2 compares the predicted spectra to the clinically measured spectra of the four tongue sites. The simulated spectra accurately predict the drop in fluorescence intensity and changes in spectral shape displayed in this example of neoplastic progression in oral mucosa [Figs. 2a, 2b, 2c, 2d].

Fig. 2

Comparison of Monte Carlo model predictions with clinical fluorescence spectra measured from four oral tongue sites in the same patient. Clinical data and predictions at $350\text{-}\mathrm{nm}$ excitation from (a) normal tongue, (b) normal tongue with severe inflammation, (c) focal mild dysplasia, and (d) moderately differentiated cancer. Bar graphs showing the predicted fraction of detected photons originating from the epithelium and the stroma for the (e) normal, (f) inflammatory, and (g) dysplastic simulations. (h) Bar graph showing the predicted fraction of detected photons originating from the cellular region (upper $400\mu \mathrm{m}$ , labeled as epithelium on the bar graph) and the deeper fibrous regions (labeled as stroma on the bar graph) for the cancer simulation.

The fraction of photons originating from the epithelium and the fraction originating from the stroma were calculated for each simulation [Figs. 2e, 2f, 2g, 2h]. (In the normal case [Fig. 2e], the detected fluorescence includes both a significant epithelial and stromal component. For example, at $450\text{-}\mathrm{nm}$ , the fraction of photons originating from the epithelium is approximately equal to the fraction of photons originating from the stroma, and the stromal contribution dominates the total fluorescence in the $400\text{-}\text{to}420\text{-}\mathrm{nm}$ region. With the presence of inflammation [Fig. 2f], the fraction of photons originating from the stroma substantially drops, while the fraction of epithelial photons increases. It should be noted that the decreased fraction of stromal photons is due to a large reduction in fluorescence originating from the stroma and not due to an increase in fluorescence originating from the epithelium. Thus the overall intensity of the predicted spectrum from the inflammation site is very low compared to the simulations from normal tissue. In the dysplastic tissue sample [Fig. 2g], the fraction of photons originating from the epithelium is very high for all emission wavelengths. Here, the high fraction of epithelial photons is due to a very large reduction of stromal fluorescence and a significant increase of epithelial fluorescence compared to the normal case. As a result, the overall fluorescence intensity of the simulated dysplastic spectrum is lower than the normal case but not as low as in the inflammation case. Thus, while both the dysplastic and inflammatory lesions show a drop in stromal fluorescence, the overall fluorescence intensity and fraction of epithelial photons of the dysplastic tissue are greater than that of the inflammatory tissue due to the increase in fluorescence originating from the dysplastic epithelium. In the cancerous tissue case [Fig. 2h], the fraction of fluorescence from the deep fibrous region is extremely low, and the overall fluorescence intensity is low as well. This is due to sharply reduced fluorescence from the deeper matrix region and slightly reduced fluorescence from the upper cellular regions of the cancer lesion relative to the analogous regions in normal tissue.

3.2.

Site-Specific Spectral Changes Caused by Variations in Keratinization

Figure 3 shows high resolution confocal images of normal oral mucosa taken from two different anatomic sites that display different epithelial fluorescence patterns. The normal buccal site [Fig. 3a] is characterized by a weakly fluorescent superficial layer, while the normal palate site [Fig. 3b] displays a brightly fluorescent superficial epithelium. Clinical fluorescence spectra at $350\text{-}\mathrm{nm}$ excitation measured from both sites are displayed in Fig. 3c.

Fig. 3

(a) and (b) Fluorescence confocal images and (c) measured fluorescence spectra at UV excitation, showing changes in autofluorescence patterns due to anatomic site variation. (a) Normal buccal mucosa; and (b) normal palate. In (a) and (b), the white lines represent the basement membrane and the yellow lines represent the tissue surface. Scale bars represent $200\mu \mathrm{m}$ . (c) Fluorescence spectra at $350\text{-}\mathrm{nm}$ excitation, measured from the same oral sites shown in the confocal images.

Figure 4 compares the Monte Carlo predictions to the clinically measured spectra of the normal buccal mucosa and normal palate sites. The fluorescence spectrum of the palate site has greater intensity and is blue shifted relative to the spectrum measured from the buccal site. These trends are accurately predicted by the model [Figs. 4a and 4b]. Predictions for the buccal mucosa site are within 10% of the measured spectra. However, the predicted intensity of the palate site is lower than that measured in vivo, possibly due to experimental error in estimating the thickness of the superficial epithelial layer. Previous model sensitivity analysis performed with the depth-sensitive Monte Carlo model has shown that even variations of $50\mu \mathrm{m}$ in the thickness of the superficial layer can result in a noticeable effect on the total intensity of the prediction.¹² The estimated thickness of the strongly fluorescent keratin layer from the confocal images has a large standard deviation, which could explain the difference in intensity between the simulated and clinical spectra.

Fig. 4

Comparison of Monte Carlo model predictions with clinical fluorescence spectra measured at different anatomic sites in the oral cavity. Clinical data and predictions at $350\text{-}\mathrm{nm}$ excitation from (a) normal buccal mucosa and (b) normal palate. Bar graphs showing the predicted fraction of detected photons originating from the epithelium and stroma for (c) normal buccal mucosa and (d) normal palate.

Table 2

Thickness, dominant fluorophore, and relative fluorescence intensity for each oral tissue site and sublayer. Values were used as input to the Monte Carlo model.

Figure 4 also summarizes the fraction of photons originating from the epithelium and the stroma derived from depth analysis of the simulated spectra. The buccal mucosa has a large fraction of photons originating from the stroma [Fig. 4c], compared to the palate site [Fig. 4d]. Here, the larger fraction of epithelial photons can be attributed mainly to increased fluorescence from the epithelial layer of the palate site. Additional analysis (data not shown) indicates that the increased epithelial fluorescence originates from the superficial, keratinized region of the epithelium in the palate site. We suggest that the presence of a brightly fluorescent keratin layer in the palate site increases the fluorescence originating from the superficial epithelium compared to the buccal site, which accounts for the higher intensity of the palate spectrum compared to the buccal spectrum. The increased contribution from keratin also affects the spectral shape of the fluorescence from the palate site. The measured spectrum of the palate site shows similarities to the emission spectrum of keratin, which displays a peak at $420\mathrm{nm}$ .¹²