2.1.

Tissue Samples

Freshly excised thick skin specimens with nonmelanoma skin cancers were obtained immediately after the Mohs micrographic surgeries performed at the Dermatologic Surgery Unit of Massachusetts General Hospital. An example digital photograph of one of the specimens is shown in Fig. 1a . Low contrast of the lesion with respect to the surrounding tissue does not allow visual demarcation of the tumor boundaries. All the experiments were conducted according to a protocol approved by the institutional review board of Massachusetts General Hospital. In total, 79 specimens with 86 tumors, including 73 BCCs and 13 SCCs were examined. The lateral size of the samples varied from 4 to $35\mathrm{mm}$ , and the thickness from 0.5 to $20\mathrm{mm}$ . The lateral size of the tumors was in the range from 0.35 to $12\mathrm{mm}$ .

Fig. 1

Photograph of a sample with nonmelanoma skin cancer (nodular BCC). The lines along the left side of the sample are surgeons’ markers that enable orientation of the excision with respect to the in situ wound. (b) Close view of the imager: 1—CCD camera; 2—CCD lens; 3—illuminator with variable angle of incidence; 4—bandpass filter and linearly polarizing filter (polarizer); 5—linearly polarizing filter (analyzer); 6—skin sample with BCC.

2.2.

Chemicals

Demeclocycline hydrochloride was purchased from Sigma Aldrich (Steinheim, Switzerland). Tetracycline hydrochloride was purchased from IVAX Pharmaceuticals, Inc. (Miami, Florida). Dulbecco’s phosphate buffered solution (DPBS, pH 7.4) was purchased from Mediatech, Inc. (Herndon, Virginia).

2.3.

Tissue Staining and Handling

The tissue was stained for approximately $5\min$ with 1 to $2\mathrm{mg}∕\mathrm{ml}$ or 0.25 to $2\mathrm{mg}∕\mathrm{ml}$ DPBS solution of TCN or DMN, respectively. To remove excess of the TCN or DMN, the specimens were briefly rinsed in DPBS. The tissue was imaged again after staining. For imaging, each sample was placed in a Petri dish on a piece of gauze soaked in DPBS and covered with a cover slip.

2.4.

Fluorescence Polarization Imaging

The fluorescence polarization imaging system is presented in Fig. 1b. Imaging was performed using the equipment and in a manner similar to that described previously.¹³ In short, TCN and DMN fluorescence was excited by linearly polarized monochromatic light centered at $390\mathrm{nm}$ with a full width at half maximum of $10\mathrm{nm}$ . Co-polarized and cross-polarized fluorescence emission images, denoted as $I_{\Vert }$ and $I_{\perp }$ , respectively, were captured by a 12-bit charge coupled device (CCD) camera in the wavelength range from 450 to $650\mathrm{nm}$ . Two sets of images of the specimens, i.e., prior ( $I_{\Vert }^{en}$ ; $I_{\perp }^{en}$ ) and following ( $I_{\Vert }^{ee}$ ; $I_{\perp }^{ee}$ ) staining, were registered. The system provided¹³ a field of view of $2.8\mathrm{cm}\times 2.5\mathrm{cm}$ , lateral resolution of $15\mathrm{\mu }\mathrm{m}$ , and axial resolution of approximately 70 to $200\mathrm{\mu }\mathrm{m}$ . The maximal integration time on the CCD array was $10\mathrm{s}$ , and the maximal incident power density on the sample was $0.08\mathrm{mW}∕{\mathrm{cm}}^2$ . All experiments were performed at room temperature.

2.5.

Data Processing

Endogenous $\left(P^{en}\right)$ and mixed $\left(P^{ee}\right)$ fluorescence polarization images (FPIs) were determined from the experimental co-polarized and cross-polarized fluorescence intensity images, acquired prior to and after tissue staining, using the definition of fluorescence polarization:

The two experimental sets of images are: ( $I_{\Vert }^{en}$ , $I_{\perp }^{en}$ )—co-polarized and cross-polarized components of endogenous fluorescence emission that were acquired prior to tissue staining, and ( $I_{\Vert }^{ee}$ , $I_{\perp }^{ee}$ )—co-polarized and cross-polarized components of the mixed (endogenous and exogenous) fluorescence emission that were acquired after tissue staining. Net exogenous fluorescence emission cannot be measured directly. However, it can be calculated as the difference of the mixed and endogenous fluorescence emission:

Substituting Eq. 2 into Eq. 1, we obtain the formula for calculating net exogenous fluorescence polarization:

2.6.

Comparison to Histopathology

Preparation of the horizontal histological frozen H&E section during Mohs surgery is described in detail elsewhere.¹⁴ As was discussed by Yaroslavsky, ¹⁵ the features of the last frozen section generated during surgery should closely resemble those of the FPI acquired from the remaining piece of tissue that we used for the experiments. Therefore, the accuracy of the fluorescence polarization imaging method was evaluated by comparison with frozen histopathology, processed during Mohs micrographic surgeries. H&E histological images and FPIs were evaluated and the tumor margins were outlined independently by two investigators capable of reading histopathology. The Mohs surgeon, who outlined the tumor in histology, did not have access to the FPIs. The investigator, who outlined the tumor in the FPIs, had no access to the histological slides. In practice, due to the preparation technique of the frozen section, it may be stretched or shrunk in comparison with the remaining thick piece of skin. To reduce the influence of these artifacts on the comparison of the results, we scanned each histopathological slide with the resolution of $15\mathrm{\mu }\mathrm{m}$ , identified four to ten pairs of common features in histology and in the corresponding FPIs, and overlaid FPIs with histopathological images by applying affine, projective, or polynomial transformations.¹⁶ These transformations are widely used for correcting artifacts of magnetic resonance imaging.^{17, 18} The accuracy of the transformations was checked by applying the algorithm for correcting the distorted image of the known object.

We have introduced the following criteria for the comparison of FPIs to corresponding histopathology. Similarity in visual appearance or the shape of tumor-affected areas in the FPIs and histology served as the first criterion.

To quantify the accuracy of the fluorescence polarization technique, the surface areas occupied by the tumor in the FPI $\left(S_{fpi}\right)$ and histological slides $\left(S_h\right)$ were processed and compared. The ratio of the cancerous areas in the FPI and histology was selected as the second criterion. We considered the agreement acceptable if the tumor area in the FPI was equal or up to 15% greater than that in histology $(1⩽S_{fpi}∕S_h<1.15)$ , i.e., would correspond to complete tumor removal by image-guided surgery. The contrast of the lesion with respect to the surrounding healthy tissue in the mixed FPI $\left(CL{N}^{ee}\right)$ was selected as the third criterion. $CL{N}^{ee}$ values greater than 0.5 were considered acceptable. This threshold was chosen to guarantee that the difference of the cancerous and normal AVFP is significantly (at least 10 times) greater than the noise level.

3.1.

Choice of the Imaging Mode

For each investigated sample, we have obtained both fluorescence emission and fluorescence polarization images. Out of 32 tumors stained in TCN and 54 tumors stained in DMN, fluorescence emission images could be used for tumor detection and delineation in only 4 and 11 cases, respectively. Thus, fluorescence emission does not provide an accurate tool for the detection and delineation of nonmelanoma skin tumors. Example fluorescence emission and polarization images of nodular BCC stained in $2\mathrm{mg}∕\mathrm{ml}$ aqueous solution of TCN are compared to corresponding H&E histopathology in Fig. 2 . In the fluorescence emission image [Fig. 2a], the tumor cannot be identified. In contrast, in the FPI, the cancerous area is bright, as compared to normal skin. The shape, size, and location of the tumor in the FPI correlate well with histopathology [Fig. 2c].

Fig. 2

Nodular BCC. Scale $\mathrm{bar}=3\mathrm{mm}$ (a) Fluorescence emission image, acquired after staining in $2\mathrm{mg}∕\mathrm{ml}$ aqueous solution of TCN, is presented. (b) Fluorescence polarization image, acquired after staining in $2\mathrm{mg}∕\mathrm{ml}$ aqueous solution of TCN, is presented. (c) Histological frozen H&E section. Tumor margins, as determined by Mohs surgeon, are outlined with black line.

3.2.

Endogenous and Exogenous Fluorescence Polarization Observed from Nonmelanoma Skin Cancer

The endogenous, mixed, and exogenous FPIs of micronodular BCC are shown in Figs. 3a, 3b, 3c , respectively. The sample was stained in $2\mathrm{mg}∕\mathrm{ml}$ aqueous solution of TCN. Bright parts correspond to the collagen fibers present in the dermis. The DMN FPI in Fig. 3b is not corrected for the endogenous fluorescence polarization. However, the tumor that displays a significantly higher level of fluorescence polarization, compared to normal skin, is much brighter than the surrounding tissue. In addition, the tumor appears as a structureless and comparatively homogeneous formation, whereas hair follicles and sebaceous glands closely resemble the appearance of the respective structures in histology. The combination of higher values of fluorescence polarization in the tumor and its characteristic morphological appearance makes discrimination of cancerous areas in the FPI possible. As a result, the tumor appears bright and can be clearly demarcated in both mixed and net exogenous FPIs. The contrast of the mixed FPI is 4.00, and the contrast of the net exogenous FPI is 3.84. The size, shape, and position of cancerous area are the same in both images and correlate well with each other and with histology, presented in Fig. 3d. The $S_{fpi}∕S_h$ ratio in this sample is 1.02. Histograms shown in Fig. 3e demonstrate that exogenous fluorescence polarization of TCN was more than 5 times higher in tumor compared to healthy skin, whereas endogenous fluorescence polarization was higher in normal tissue. This representative case demonstrates that endogenous fluorescence polarization is much lower than exogenous and that cancerous areas in the sample can be accurately demarcated in the mixed FPI. There are no significant differences in the contrast, shape, and size of the tumors in the net exogenous and mixed FPIs.

Fig. 3

Nodular BCC. Scale $\mathrm{bar}=2\mathrm{mm}$ . (a) shows endogenous fluorescence polarization image prior to tissue staining. In (b), mixed exogenous and endogenous FPI, acquired after staining in $2\mathrm{mg}∕\mathrm{ml}$ aqueous solution of TCN, is presented. (c) Demonstrates the pure exogeneous FPI of TCN that was obtained using the images shown in (a) and (b). (d) Histological frozen H&E section. Tumor margins, as determined by Mohs surgeon, are outlined with black line. (e) Average fluorescence polarization measured in normal and cancerous areas of the nodular BCC from (a), (b), and (c). Bars are standard errors.

3.3.

Exogenous Fluorescence Polarization Examination of Nonmelanoma Skin Cancers

The potential of the fluorescence polarization imaging technique was evaluated by comparing the imaging results to histopathology of the respective tumors. To mimic the clinical conditions, in the following figures, we present and compare to histology only the mixed FPIs that were not deconvolved using previously acquired endogenous FPIs. The mixed FPIs are dominated by the signal from the exogenous tetracyclines; therefore, in the remainder of this paper, these images are referred to as exogenous FPIs.

All FPIs were diagnosed independently of the histologic images. In Figs. 2, 3, 4, 5 , the images of the most common types of nonmelanoma skin cancers are presented side by side with corresponding histopathology.

Fig. 4

Infiltrative BCC. Scale $\mathrm{bar}=2\mathrm{mm}$ . Comparison of (a) fluorescence polarization image (stain: $0.5\mathrm{mg}∕\mathrm{ml}$ aqueous solution of DMN) with (b) the last frozen histological H&E section of the same specimen. Tumor margins, as determined by Mohs surgeon, are outlined with black line. In (c), average values of fluorescence polarization in cancerous and normal tissues determined from FPI presented in (a) are shown. Bars are standard errors.

Fig. 5

Invasive SCC. Scale $\mathrm{bar}=1\mathrm{mm}$ . Comparison of (a) fluorescence polarization image (stain: $0.5\mathrm{mg}∕\mathrm{ml}$ aqueous solution of DMN) with (b) the last frozen histological H&E section of the same specimen. Tumor margins, as determined by Mohs surgeon, are outlined with black line. In (c), average values of fluorescence polarization in cancerous and normal tissues determined from FPI presented in (a) are shown. Bars are standard errors.

In Figs. 2b and 3b, the FPIs of the most frequently occurring type of nonmelanoma skin cancer, nodular BCCs, are shown. This prototype of BCC consists of nodular solid aggregates of tumor in the dermis, usually arising from the epidermis. Clinically, it may have a very subtle contrast with imprecise margins. The tumors presented in Figs. 2 and 3 were stained with $0.25\mathrm{mg}∕\mathrm{ml}$ and $2\mathrm{mg}∕\mathrm{ml}$ aqueous solutions of DMN and TCN, respectively. In both cases [Figs. 2b and 3b], the tumors have excellent contrast with respect to the surrounding tissue. The location, size, and shape of the tumors in the FPIs correlate well with corresponding histopathology.

In Fig. 4, infiltrative BCC stained with $0.5\mathrm{mg}∕\mathrm{ml}$ aqueous solution of DMN is shown. Of all types of BCC, this kind of tumor is the most aggressive with the highest recurrence rate. Structurally, it is composed of the thin cords of tumor cells surrounded by normal skin tissues, such as collagen. As a result, it has a very weak contrast with respect to the surrounding healthy skin, both visually and in histology [see Fig. 4b]. In the FPI that is presented in Fig. 6a , the tumor is very bright, with an average value of fluorescence polarization of $3.17\times {10}^{-2}$ , whereas the AVFP of normal tissue was $1.14\times {10}^{-2}$ , which corresponds to a $CL{N}^{ee}$ of 2.03 [see Fig. 4c]. Comparison of the FPI and histology indicates excellent correlation. The ratio of $S_{fpi}$ to $S_h$ in this sample is 1.05.

Fig. 6

Effect of DMN concentration on the contrast of cancer with respect to healthy tissue. Bars are standard errors. (b) Effect of TCN concentration on the contrast of cancer with respect to healthy tissue. Bars are standard errors.

In Fig. 5, we present an example image of invasive SCC. SCCs often exhibit an unpredictable growth pattern and a generally higher potential for metastasis as compared to BCCs. Unlike BCC, which originates from the basal cell layer, SCC originates from keratinocytes and therefore has a different morphological appearance and staining pattern. Comparison of the FPI and histology, presented in Fig. 5a and 5b, respectively, shows good correlation $(S_{fpi}∕S_h=1.06)$ . The histograms given in Fig. 5c demonstrate that the values for AVFP for the SCC tumor are much higher than in the adjacent healthy skin, with the $CL{N}^{ee}=3.02$ .

In total, we have imaged 79 specimens with 86 tumors. Out of 86 tumors, 13 were invasive SCCs and 73 were BCCs, including nodular, micronodular, infiltrative, and superficial subtypes. This is a fair sampling, as SCCs account for approximately 20% of all nonmelanoma skin cancers. Forty-nine specimens with 54 tumors were stained using DMN. In 51 cases, the location, shape, and size of the tumors were identified correctly. One superficial BCC tumor stained in $0.5\mathrm{mg}∕\mathrm{ml}$ DMN solution was missed. In two cases, we have identified two false positive tumor nests in two different samples (an invasive SCC and a superficial BCC, both stained in $0.25\mathrm{mg}∕\mathrm{ml}$ DMN solution). Thirty specimens with 32 tumors were stained using TCN. In 28 cases, the location, shape, and size of the tumors were identified within the criteria specified in Sec. 2.6. In two cases, the location of the tumor lobules (both were nodular BCCs) were determined correctly, but the ratios of the surface areas occupied by the tumor in the FPI and histological slides, $S_{fpi}∕S_h$ , was greater than 1.15. In one case of nodular BCC, the tumor was missed, and in one case of nodular BCC, a false positive tumor nodule was found in the FPI.

3.4.

Effect of Stain Concentration on Fluorescence Polarization Imaging

We have evaluated several concentrations of the fluorophores for tissue staining. Concentrations of $2\mathrm{mg}∕\mathrm{ml}$ (7 samples), $1\mathrm{mg}∕\mathrm{ml}$ (8 samples), $0.5\mathrm{mg}∕\mathrm{ml}$ (14 samples), and $0.25\mathrm{mg}∕\mathrm{ml}$ (18 samples) were tested for DMN. The results, presented in Fig. 6a, show that the average values of DMN fluorescence polarization in the tumors are higher than in the surrounding healthy skin for all the investigated concentrations of the fluorophore. The averaged over the samples contrast of the lesions with respect to normal skin, $CL{N}^{ee}$ , was found to be equal to 2.03, 2.31, 1.40, and 1.94 for the DMN concentrations of $0.25\mathrm{mg}∕\mathrm{ml}$ , $0.5\mathrm{mg}∕\mathrm{ml}$ , $1\mathrm{mg}∕\mathrm{ml}$ , and $2\mathrm{mg}∕\mathrm{ml}$ , respectively. The data were statistically analyzed using Student’s t-test for the paired populations. The differences between cancerous and normal tissues were significant for all the concentrations of the stain $(p<0.001)$ .

For the TCN, we have used the concentrations of $2\mathrm{mg}∕\mathrm{ml}$ and $1\mathrm{mg}∕\mathrm{ml}$ only. The employed excitation wavelength centered at $390\mathrm{nm}$ is in the vicinity of the absorption maximum of DMN, whereas the absorption band of TCN is shifted considerably toward the shorter wavelengths. Excitation of TCN fluorescence using wavelengths shorter than $390\mathrm{nm}$ was found to be inefficient due to the lower throughput of the system in the UV and considerably higher attenuation in skin below $390\mathrm{nm}$ . Comparison of the results obtained using different concentrations of TCN is presented in Fig. 6b. As in the case of DMN, the average fluorescence polarization of TCN was found to be significantly higher in cancerous as compared to normal skin for both concentrations of the fluorophore. The averaged over the samples contrast of the lesions with respect to normal surrounding tissue, $CL{N}^{ee}$ , was found to be equal to 2.70 and 2.36 for the TCN concentrations of $1\mathrm{mg}∕\mathrm{ml}$ and $2\mathrm{mg}∕\mathrm{ml}$ , respectively. Statistical analysis of the data using Student’s t-test for the paired populations has shown that the differences between normal and cancerous tissues were significant for both concentrations of the stain $(p<0.001)$ .