1.

Introduction

Melanoma is a malignant tumor derived from melanocytes, which are found predominantly in the skin and are responsible for the production of the dark pigment melanin. It is more common in fair-skinned populations and is responsible for the majority of skin-cancer-related deaths.^1–4 The clinical progression of the patients suffering from skin melanoma is directly associated with the early diagnosis of the tumor. Although the prognosis of patients with early thin melanomas is favorable after sufficient surgical excision, when metastases occur, melanoma is incurable with high mortality rate.^2,5

Dysplastic nevi are benign melanocytic nevi, which are clinically atypical with irregular outline and a variegated appearance and are histologically characterized by architectural disorder and cytological atypia. Their significance in relation to melanoma is high since they can be considered stimulants of melanoma and markers of individuals at increased risk of developing melanoma. They are also potentially and occasionally actual precursors of melanoma. Morphologically and biologically, they are intermediate between common nevi and melanoma. The presence of dysplastic nevi in patients may mean deficient DNA repair and association with overexpression of pheomelanin, which may lead to increased oxidative damage and increased danger for DNA damage and tumor progression.^1,6–10

The discrimination between malignant and benign melanocytic lesions is one of the major issues of clinical dermatology. Clinical studies report that the diagnosis by simple inspection has sensitivity 65 to 80%, depending on dermatologist experience, whereas dermoscopy increases this rate by approximately 20%, but is also dependent on the experience and training of the clinician.^11–15 In addition, the assessment of the depth of invasive melanoma is an important issue of clinical dermatology and essential for the determination of the prognosis and the selection of the suitable treatment. The necessity for improvement of the diagnostic accuracy of melanocytic lesions led to the development of imaging methods such as in vivo confocal laser scanning microscopy,^16–25 high-resolution ultrasound (U/S),^26–29 spectroscopy,^30–33 and optical coherence tomography (OCT),^29,34–37 but none of them has gained wide acceptance as a diagnostic tool so far. Recently, an optical method using second harmonic generation imaging of collagen was investigated for demarcation of melanoma borders, without offering detailed and accurate evidence for the discrimination of the malignant from benign melanocytic lesions.³⁸

Herein, we present our initial results of a novel, non-contact diagnostic technique for three-dimensional (3-D) imaging of melanocytic lesions, using optical computed tomography (optical-CT), in order to discriminate melanoma from benign nevi and determine the thickness of invasive melanoma.³⁹ This method is based on non-contact measurements of biopsy samples using diffused light illumination and multiple angle detection of transmitted light.^40,41 A filtered back-projection algorithm was employed to calculate 3-D maps of the optical properties of the tissue samples. These maps were used for determining the depth and the type of lesion, with high accuracy and compared on a double-blind basis with histology. Based on the difference of the obtained optical properties, benign from malignant lesions could be consistently differentiated. To our knowledge, this is the first time that an appropriately adapted optical-CT method is applied to excised tissue samples without optical clearing or any kind of chemical fixation or alteration of the biopsies, in contrast with previous work.^42–44 The value of developing such a method could be high for the treatment of melanoma, assisting the physician for a rapid evaluation of a melanocytic lesion and of its thickness. The treatment would be finalized after definitive histological results.⁴⁵

2.

Materials and Methods

2.1.

Experimental Setup

The optical-CT system developed for this study provides non-contact multiple-angle measurements of biopsy samples in order to calculate 3-D maps of their optical properties.^46,47 The principle of the system is depicted schematically in Fig. 1. The light source was a white light lamp consisting of 80 light-emitting diode (LEDs). A diffuser was placed between the light source and the optical components directing the light onto the sample. Diffuse light was preferred rather than collimated laser beam for homogeneous illumination and for acquiring average values of the optical properties. The light was then reflected by a flat mirror, and it was guided to a linear polarizer (Hoya, Japan). After transmission through the sample, which was positioned on a rotational stage (Standa, Lithuania) covering 360 deg of rotation, the linear polarization of light was mostly lost due to scattering. Ambient light was blocked by a black, totally absorbing, paper tube. Light signals were detected by a sensitive charge coupled device camera (CCD-Camera) (Andor Corp., DV434, Belfast, Northern Ireland), which was thermoelectrically cooled down to $-20°\mathrm{C}$. The CCD-Camera was equipped with a SIGMA 50 mm $f/2.8$ objective lens (SIGMA Corporation, Tokyo, Japan). A second polarizer (Hoya, Japan) was attached to the input of the lens and was adjusted either parallel or perpendicular to the polarization of the light used for illumination. The whole experimental procedure was controlled with in-house-developed software, and images were stored on a PC.

Fig. 1

Schematic representation of the optical-CT setup (LED: light-emitting diode; CCD: charge coupled device).

3.

Results

The study was carried out on a total of 16 samples, divided as follows: six samples were identified as malignant melanomas of the skin, four as dysplastic melanocytic nevi, and six were histologically characterized as nondysplastic melanocytic nevi (however, these had been characterized clinically and dermoscopically as atypical). One common nevus, one hemangioma, and normal skin were used as controls. The highest values of the AC obtained from the fits described in Sec. 2.4 for each sample are presented in Table 1.

Table 1

Highest attenuation coefficient of different melanocytic lesions and their estimated thickness assessed by LED optical tomography and conventional pathological examination. The values were rounded to the closest significant figure.

In Fig. 2, three characteristic melanocytic lesions are presented: one nondysplastic nevus (2.1), one dysplastic nevus (2.2), and one melanoma (2.3). The images of columns A, B, C, and D correspond with simple white-light photographs, histological images, 3-D reconstructions, and AC plots as a function of depth, for the related samples. From the plots of the AC against depth, it is observed that the highest value obtained was $7.20\pm 0.07{\mathrm{cm}}^{-1}$, and it corresponded to melanoma (2.3.D). In the dysplastic nevus the highest AC value was $5.89\pm 0.15{\mathrm{cm}}^{-1}$ and in nondysplastic nevus $3.14\pm 0.13{\mathrm{cm}}^{-1}$. As shown also in Table 1, in which the findings for each lesion are presented, the highest AC values correspond with the malignant lesions, where morphological atypia of the cells and architectural disorder of the tissue are more intense. We attribute the increased AC to the higher scattering properties of dysplastic and malignant tissue. Hence this approach provides the ability to evaluate each pigmented skin lesion according to its cytological atypia, which is characteristic for its malignancy.

Fig. 2

Characteristic images of lesions from each studied group: (2.1) nondysplastic nevus; (2.2) dysplastic nevus; (2.3) melanoma. (a) Macroscopic image of the lesion. (b) Histological image of the lesion with measured depth. (c) Image of an axial slice obtained from the 3-D reconstructed image. The grey line shows the location of the profile. (d) Plot of the attenuation coefficient against depth obtained along the grey line of image 2C.

In Fig. 3, the plots of the attenuation coefficient against depth obtained from: (a) normal skin; (b) hemangioma; and (c) common nevus are presented. The plots in (d); (e); and (f) show the same profiles plotted with the scale of the graphs of Fig. 2 for a direct comparison. As can be noticed, the highest value of attenuation coefficient in these three plots is $1.47{\mathrm{cm}}^{-1}$, and it is obtained in the common nevus. The corresponding value for hemangioma is $0.68{\mathrm{cm}}^{-1}$ and for normal skin $0.46{\mathrm{cm}}^{-1}$. These three samples were used as controls of non-atypical tissue. The highest values of attenuation coefficient in these samples are much lower than those in cutaneous melanomas, in dysplastic nevi, and even in nondysplastic nevi. This is due to the lower concentration of pigments, even for common nevus and to the fact that there is no morphological atypia of the cells and architectural disorder of the tissue. Therefore, the attenuation coefficient values are very low, and these samples can be easily differentiated from atypical or malignant lesions. This constitutes a confirmation of our results and our suggestions, whereas samples of normal skin or benign lesions of the skin have low values of AC and can be easily evaluated and discriminated from atypical and malignant pigmented skin lesions.

Fig. 3

Plots of the attenuation coefficient against depth obtained from: (a) normal skin; (b) hemangioma; and (c) common nevus. The plots in (d), (e), and (f) show the same profiles plotted with the scale of the graphs of Fig. 2 for a direct comparison.

Determination of the thickness of cutaneous melanoma is important for assessment of further therapy of the pathological condition. Figure 4 depicts the effective depth calculated with the fitting process described in Sec. 2.4, plotted against the thickness provided by standard pathological assessment for each sample. The data were fitted with a linear formula with an $R^2$ value of 0.987, highlighting the excellent agreement of the optical-CT and histology results. No statistically significant difference was observed when thickness values obtained by optical-CT were compared with histological thickness using a Wilcoxon signed rank test ($P=0.243$). The absolute difference of histological thickness minus AC estimated thickness ranged from 0.02 to 0.30 mm (median: 0.05 mm), exceeding 0.1 mm only in one sample (No. 3, see Table 1). The assessment of this thickness determines the prognosis and the selection of the optimal treatment, which is important for the physician, assisting him in a much shorter time than histology to decide and choose the most favorable management of melanoma.⁴⁵ At this point we need to emphasize that the investigation was doubly blind, meaning that the involved parties had no information regarding the results from each other.

Fig. 4

Thickness estimated with the use of attenuation coefficient (AC) plotted against thickness provided by standard pathological assessment for each sample.

Finally, as presented in Fig. 5, there was a statistically significant difference of highest AC values between malignant melanomas, dysplastic nevi, and nondysplastic nevi as indicated by Kruskal-Wallis test ($P=0.001$). Median (range) AC values were higher for malignant melanomas compared with dysplastic and nondysplastic nevi [7.67(0.86) vs. 5.63(0.35) vs. 3.45(2.05), respectively]. The range of the value for the nondysplastic nevi appears to be higher only because we have included in this category the one common nevus sample, the value of which is presented by the asterisk in the graph of Fig. 5. These results suggest that optical-CT can be employed for the characterization and differentiation of pigmented skin lesions and assist the immediate evaluation of biopsy samples, based on the 3-D reconstructed maps of the total attenuation coefficient.

Fig. 5

Box plot of highest values of attenuation coefficient (AC) for each group of lesions ($P<0.01$). The asterisk represents the value of the one common nevus sample.

4.

Discussion

In this paper we have demonstrated that optical-CT can be a valuable tool for imaging and characterization of biopsy samples of melanocytic tumors. Our study provides preliminary evidence that the 3-D maps of attenuation coefficient as provided by optical-CT may be used for the classification of common nevi, dysplastic nevi and cutaneous melanoma, as well as for the evaluation of the maximum thickness including Breslow thickness for melanomas.

The diagnostic accuracy of melanocytic lesions is an important factor for the prognosis and therapeutic interventions of skin cancer. This has created the need for the development of imaging methods such as in vivo confocal laser scanning microscopy,^16–25 high-resolution ultrasound (U/S),^26–29 spectroscopy,^30–33 and OCT.^29,34–37 However, none of them to date has been established in the diagnosis of melanoma.

OCT is an imaging modality, which provides high-resolution cross-sectional images. OCT is equivalent to confocal microscopy but with an added coherence gate, which decreases the noise floor. For both techniques, the contrast mechanism is backscattered light reflection. Similarly, ultrasound imaging uses backscattered sound waves for detecting the echo reflected from different tissue layers. It is a method that has been used to visualize and characterize melanocytic skin lesions. Related studies detected significant differences between benign nevi and malignant melanomas and suggested features that may serve in differentiation of melanocytic lesions. Although OCT appeared as a promising noninvasive imaging technique for the micromorphology of the skin, it has not been studied systematically in malignant melanoma (MM), and its major limitation is penetration depth, which cannot exceed 1 mm.^29,34–37

Recently, second harmonic generation has been demonstrated for imaging the collagen on skin lesions and determining melanoma borders, based on collagen morphology and density. However, despite the high sophistication of the technique, it did not offer detailed and accurate evidence for the discrimination of the malignant from benign melanocytic lesions, yet.³⁸

In this work we have chosen to explore the capabilities of optical-CT with in-house-developed system using white-light illumination with LEDs, due to the 3-D attenuation maps it can deliver. A diffused light illumination pattern was preferred exploiting 80 white LED lights, to achieve a uniform irradiation of the whole sample. A rotation stage with a purpose build sample holder was used to cover 360 deg of rotation, with a CCD capturing images every 5 deg, ensuring that the entire sample was illuminated and transmitted light signals were detected from several angles. An important feature of the system was the use of two polarizers, enabling to distinguish between light transmitted through or around the samples. This resulted in an increase in sensitivity and dynamic range of scattered light detection and allowed the imaging of thick biopsy samples. In this study only the perpendicular polarization measurements were used to reconstruct the 3-D image for each sample. The parallel polarization could instead be used for reconstructing the surface boundaries of the sample and for separating scattering from absorption using the radiative transfer equation with appropriate phase functions such as the Rayleigh phase matrix.^53,54 Our initial experiments were accomplished ex vivo, as an optical 3-D histology, in order to evaluate different types of tissue and study the light transport through macroscopic samples. In the initial investigations a simple back-projection reconstruction algorithm provided the 3-D map of the attenuation coefficient, which enabled us to evaluate the melanocytic lesions in their whole thickness.

The optical-CT reconstructions provide 3-D maps of the total attenuation coefficient inside the biopsy samples, which correspond with the sum of absorption and scattering coefficients. Absorption in the samples investigated can be mainly attributed to melanin, which is the main absorber of melanocytic lesions. Thus the reconstruction provides an image of the differences in the distribution and quantity of melanin at different areas of the tissue. However, the variation in measured optical properties between nevi and melanoma may not be totally explained by differences in the melanin content of the lesions, as melanin content may be similar in different types of melanocytic lesions. An important difference between benign and malignant lesions is attributed to the morphological atypia of the cells and architectural disorder of the tissue of the malignant lesions, which can influence (increase) light scattering, thus affecting the total attenuation coefficient, on which our results are based. Hence a higher total attenuation coefficient can be expected from the malignant lesions, which is what we have also measured with our optical-CT system. The ability to discriminate benign from malignant melanocytic lesions is thus based on the differences in the measured attenuation coefficient in different types of melanocytic lesions, which depend on the atypia of cancer cells. In addition, the different values of attenuation coefficient in different tissues, as well as in different cells, concerning their morphological formality offer us the ability to detect the boundary of the lesion and determine the thickness of invasive melanoma.

At this point it must be noted that although our study provided remarkably accurate results, it has been performed with a simple back projection algorithm, which does not account for light diffusion in tissue. This was compensated partially by the use of the diffuser and the polarizers, as described in detail in the Sec. 2. We are, however, researching the development of algorithms that use the diffusion theory for the reconstructions and will be able to dynamically adjust the phase function, which describes the scattering. These approaches include:

Furthermore, sensitivity and specificity studies including other skin tumors [basal cell carcinoma (BCC); squamous cell carcinoma (SCC)] except the melanocytic lesions described in this paper are under way.

Provided that the results of the current study will be confirmed by larger scale studies then optical-CT might allow for an immediate examination of the biopsies, ideally assisting the physician to have a direct classification of the lesion along with an evaluation of its thickness in a much shorter time than via histology. The assessment of this thickness can determine the prognosis and the selection of the most favorable management of melanoma.⁴⁵ This will lead to lower cost for the health system and less inconvenience for the patient.