2.1.

Subjects

In this study, the inclusion criteria were patients aged 18 to 90 years and having pigmented skin lesions that would undergo surgical excision. The patients who had infectious skin disease on the examination sites were excluded. HGM was performed ex vivo on surgical samples and in vivo directly on the lesional skin before surgical excision. After HGM observation, all surgically removed specimens were sent for histopathological analysis to confirm the diagnosis. Among them, there were 47 nonmelanoma pigmented lesions, but HGM image acquisition from five in vivo lesions (two melanocytic nevi, two BCC, and one SK) could not be completed because of an overlying thick horny layer (e.g., nail and palm) or lesions being located on areas with irregular curvature that the objective lens could not closely approach (e.g., the nasal bridge). HGM images of 42 nonmelanoma pigmented lesions (14 ex vivo and 28 in vivo) from 40 patients (24 men and 16 women, age range of 23 to 82; two patients had two lesions), which included 18 melanocytic nevi (2 ex vivo and 16 in vivo), 10 SK (2 ex vivo and 8 in vivo), and 14 pigmented BCC (10 ex vivo and 4 in vivo; nodular and superficial types), were evaluated in this study. The process of image acquisition was performed under the protocol reviewed and approved by the Research Ethics Committee of National Taiwan University Hospital. Informed consent was obtained from each subject prior to study entry.

2.2.

Harmonic Generation Microscopy

The HGM system was modified from a commercial scanning system (Olympus, FV300) and was excited by a Cr:forsterite laser with a wavelength of 1230 nm, a pulse width of 100 fs, a repetition rate of 110 MHz, and an output average power of 500 mW. This excitation wavelength enables minimal light attenuation in skin and mucosa tissues.³³ Figure 1 is the schematic diagram of the HGM imaging system. The collimated laser beam transmitted through the scanning system, an 865-nm dichroic beam splitter (DBS1), and an infrared water immersion objective (Olympus, $\mathrm{UPlanApo}/60\times /\mathrm{NA}=1.2$) to excite skin or skin samples. The backward harmonic generation signals were collected by the same objective and reflected by DBS1 to two photomultiplier tubes (Harmamatsu R4220P for THG and Harmamatsu R928P for SHG). SHG and THG signals were divided by a 490-nm dichroic beam splitter (DBS2) and were filtered by two bandpass filters (D410/30 for THG and D615/10 for SHG) inserted. Based on the Cr:forsterite laser at 1230 nm, our previous HGM studies have indicated a diminished two-photon fluorescence in viable epidermis and dermis by measuring the spectra in live human skin.^23,34,35 Submicron spatial resolution could be achieved (lateral $<0.5\mu \text{m}$ in superficial layers and $<0.7\mu \text{m}$ at a 270 μm depth) for THG, which is the primary contrast modality for cancer diagnosis in this study. The total exposure time of the laser light for one volunteer was equal to or slightly less than 30 min. The average excitation power after the objective was around but slightly less than 100 mW. The accumulated photon energy was $\sim 180\mathrm{J}$ in each volunteer. Under such an accumulated light dose, no erythema, pigmentation, or blister formation on the examined skin was found in this study. For all followed-up histological examinations by pathologists on the illuminated specimens, no evidence of photodamage, such as coagulation necrosis, was found.

Fig. 1

Harmonic generation microscopy imaging system. The illumination source was a Cr:forsterite laser. The collimated laser beam performed a scanning system. Second-harmonic-generation and third-harmonic-generation (THG) signals were divided by a dichroic beam splitter and guided to two photomultiplier tubes with bandpass filters inserted.

2.3.

HGM Feature Description

Under THG images, the cell nuclei appear dark and the cytoplasms of cells appear bright. The proliferation of cells means an increase in cell number. According to the uniformity of cellular size and morphology, the proliferation of THG-bright cells with bright cytoplasm and dark nucleus in the stratum granulosum and spinosum was defined as monomorphous or polymorphous cells. The cell nests in rete ridges or in the dermis were also recognized as two types, namely monomorphous and polymorphous cell nests. Elongation of rete ridges represented the presence of rete ridges that could be observed continuously for a depth beyond 50 μm. Normal epidermal stratification indicated that the stratum corneum, granulosum, and spinosum could be visualized clearly in serial optical sections. Acanthotic epidermis meant that the thickness of epidermis was increased compared to the surrounding normal skin. Changes in connective tissues represented altered distribution or density of SHG-bright collagen fibers.

2.4.

Image Analysis and Statistics

The database of HGM images from 42 nonmelanoma pigmented tumors was retrospectively analyzed by three independent observers in a blinded fashion (blinded to participant name, sex, age, and diagnosis). One week before the assessment, the observers were instructed in the interpretation of the en face HGM images using representative images for each HGM criterion. During the assessment, the presence or absence of the HGM criteria (Table 1) was documented, and a tentative diagnosis was given under observer’s general assessment for individual case. The HGM data were compared with the histopathological findings that were exploited as the gold standard. Sensitivity and specificity analysis was performed with cross-tabulations and SPSS 12.0 software (SPSS Inc., Chicago, Illinois). Calculations were carried out for each criterion alone and in combination with two major criteria.

Table 1

Sensitivity and specificity of individual harmonic generation microscopy (HGM) diagnostic criterion.

3.1.

HGM Images of Healthy Skin

Morphologic information of normal skin was assessed in vivo according to our previous HGM studies.^19,30,31 As shown in Figs. 2(a) to 2(d), different cell morphologies and sizes can be visualized in optical sections taken at different depths parallel to the skin surface through THG images (presented by purple pseudocolor). Due to the multilayer structures of the stratum corneum and lipids within the corneocytes, the stratum corneum appeared bright with a strong contrast of THG [Fig. 2(a)]. As the imaging depth moved deeper into the viable epidermis [Figs. 2(b) to 2(d)], the stratum granulosum, spinosum, and basale could be visualized showing cells with bright cytoplasm and dark nuclei. The sizes of basal cells are smaller than the cells in the spinous layer, and the cytoplasm of basal cells revealed strong THG contrasts owing to the resonance enhancement of melanin.¹⁹ In addition to cell morphology, HGM provided the contrast of collagen fibers from SHG (presented by green pseudocolor) in the dermis [Fig. 2(f)]. Thus, the dermo-epidermal junction [Fig. 2(e)] could be explicitly observed by combining SHG and THG signals.

Fig. 2

The harmonic generation microscopy (HGM) images of normal skin on arm. A representative in vivo series of horizontal HGM images from healthy skin at different depths relative to the surface (5, 15, 30, 45, 60, and 85 μm) showed different epidermal layers [(a) to (d)], the dermo-epidermal junction (e), and the dermis (f). SC, stratum corneum; SG, stratum granulosum; SS, stratum spinosum; SB, stratum basale. $\mathrm{Bars}=20\mu \text{m}$.

3.2.

HGM Images of Melanocytic Nevus

Normal epidermal stratification beginning from the stratum corneum, granulosum to spinosum was observed. The presence of the aggregation of THG-bright cells in the dermo-epidermal junction [Fig. 3(a), dashed circle] and in the dermis [Fig. 3(c), arrows] was found in junctional and intradermal melanocytic nevus, respectively, in accordance with the nests of nevomelanocytes found by the conventional pathologic examination from the same tumors [Figs. 3(b) and 3(d)]. The cells in the nests were monomorphic in size and shape, and could be revealed from both THG imaging and the histopathologic examination. The THG-bright cells distributed linearly along the elongated rete ridges represented basal hyperpigmentation in junctional nevus. Some scattered melanocytes [Fig. 3(a), arrow], which had dendritic processes and were larger than basal cells, were also found in the stratum basale.

Fig. 3

Representative in vivo HGM images of junctional nevus and intradermal nevus. (a) THG-bright monomorphous nevus cell nests (dashed circle) and scattered dendritic melanocytes (arrows) in the elongated rete ridges were observed in a junctional nevus. The junctional nests can be seen as a compact, well-outlined structure in contiguity with the basal layer. (c) Well-circumscribed, THG-bright monomorphous cell nests (arrows) in the dermis of an intradermal nevus. Comparative hematoxylin and eosin (H&E)-stained skin biopsy of the junctional nevus (b) and the intradermal nevus (d), respectively (vertical plane). $\mathrm{Bars}=20\mu \text{m}$.

3.3.

HGM Images of Acanthotic Seborrheic Keratosis

The series of in vivo HGM images obtained from an acanthotic SK at different depths [Figs. 4(a) to 4(f)] showed the proliferation of THG-bright keratinocytes in the lower epidermis. Due to the increased melanin content, THG-bright cytoplasm was observed in some spinous cells and the majority of monomorphous basaloid cells [Fig. 4(c), arrows]. More epidermal HGM sections were taken in SK than that in normal skin, which indicated epidermal acanthosis. Intact dermo-epidermal junction with linearly aligned normal basal cells could be seen clearly [Fig. 4(f), arrows]. The corresponding histological section showed acanthotic epidermis and proliferation of melanin-containing basaloid cells [Fig. 4(g)].

Fig. 4

Acanthotic seborrheic keratosis. [(a) to (f)] A representative series of in vivo HGM images at different depths relative to the surface (5, 40, 65, 85, 105, and 130 μm). The whole thickness of the epidermis increased to $\sim 130\mu \text{m}$. (a) and (b) Thickened stratum corneum ($\sim 40\mu \text{m}$) was observed, which correlated with the hyperkeratosis of seborrheic keratosis. (c) Scattered THG-bright spinous cells were noted (arrows), which resulted from melanin retention in the cytosol. (d) to (f) Proliferation of monomorphous THG-bright basaloid cells in the lower epidermis. The dermo-epidermal junction was intact with linearly aligned basal cells noted [arrows; (f)]. (g) H&E-stained section of the lesion. $\mathrm{Bars}=20\mu \text{m}$.

3.4.

HGM Images of Basal Cell Carcinoma

Many characteristic morphological abnormities were present in BCC through in vivo HGM observations. There were THG-bright cell islands raising from the stratum basale and extending into the dermis [Figs. 5(d) to 5(f)]. These THG-bright cells appeared polymorphous with variation in size and shape. In contrast to the round to oval shapes of the normal basal cells with well-defined borders, many of the BCC cells became elongated with irregular borders. The presence of peripheral palisading cells in the tumor nodules was detected in vivo by HGM [Fig. 5(e), dashed line]. Collagen bundles surrounding the tumor nests [Fig. 5(e), dashed circle] in the reticular dermis became more thickened and coarse than the normal counterpart as revealed from SHG images [Fig. 5(e), yellow arrow]. The most striking finding was the presence of a large number of cells with dendritic processes [Figs. 5(f) and 5(g) to 5(i), arrows].

Fig. 5

Basal cell carcinoma. (a) to (f) A representative series of in vivo HGM images of a case of basal cell carcinoma (BCC) at different depths relative to the surface (5, 25, 40, 60, 75, and 90 μm) revealed proliferation of polymorphous THG-bright tumor cells from the stratum basale (d) and continuously extended into the dermis [dashed circle; (e)] from one BCC patient. Several tumor nests [arrow; (f)] were found in the dermis. The tumor cells appeared irregular, elongated, and peripheral palisaded [dashed line; (e)]. Collagen bundles [yellow arrow; (e)] surrounding the tumor nests were more thickened and coarse than the normal counterpart. A characteristic feature noted was the presence of a large number of cells with dendritic processes [arrow; (f)], and the feature can also be found from the other patient with deeply invasive BCC at different depths relative to the surface (80, 150, and 270 μm) [arrows; (g) to (i)]. (j) H&E-stained section of the first lesion. SC, stratum corneum; SG, stratum granulosum; SS, stratum spinosum. $\mathrm{Bars}=50\mu \text{m}$.

3.5.

Sensitivity and Specificity of HGM Criteria

The images from 42 tumors were analyzed in an observer-blinded manner to evaluate the presence or absence of HGM diagnostic morphologic criteria, which were established according to the traditional histopathological features of BCC, melanocytic nevus, and SK. Using the corresponding pathologic examination of each lesion as the golden reference, Table 1 shows summaries of the sensitivity and specificity for each HGM diagnostic criterion. We selected two HGM criteria with the highest sensitivity and specificity as major criteria for each disease. Take BCC, for instance; the presence of polymorphous basaloid cells and cells with peripheral palisading were selected as major HGM criteria for diagnosing BCC because they were the two criteria with the highest sensitivity and specificity compared to the other two. As mentioned previously, the dendritic cell can be commonly found in BCC cases. We obtained the sensitivity and specificity of the dendritic cells as 95% (95% CI: 72 to 99%) and 89% (95% CI: 70 to 94%), respectively. As the presence of dendritic cells is not a diagnostic gold standard in pathology, this feature is not included in the statistics for diagnostic performance.

3.6.

Sensitivity and Specificity of HGM for Diagnostic Performance

Our aim is to explore the clinical applicability of HGM by evaluating the diagnostic performance of HGM for differential diagnosis of nonmelanoma pigmented skin tumors. We analyzed the diagnostic performance from 42 pigmented tumors through the direct general assessment (overall evaluation) or the presence of two imaging criteria (major criteria evaluation). In the overall evaluation, each specified HGM criterion was taken into account under the diagnostic decision process, and the observers were asked to give the final diagnosis under general assessment. Based on a judgment process similar to the traditional pathological interpretation, 92% sensitivity (95% CI: 67 to 97%) and 96% specificity (95% CI: 83 to 99%) could be achieved (Table 2). In major criteria evaluation, which exploited the two specified criteria with the highest sensitivity and specificity for the diagnostic decision, 94% sensitivity (95% CI: 70 to 99%) and 100% specificity (95% CI: 87 to 100%) were achieved (Table 2). Taken together, excellent diagnostic performances with high sensitivity and specificity can be achieved through direct general assessment or the presence of two imaging criteria. These results indicate that HGM can provide real-time, accurate diagnostic outcomes. It is important to note that not all considered features are present in one lesion, which is true for our in vivo images and for traditional stained images. In addition, we also evaluated the performance of HGM for differential diagnosis in direct general assessment by comparing the ex vivo and the in vivo data. The sensitivities for in vivo data and ex vivo data were 92 and 99%, respectively. The specificity of 95% for in vivo data and 96% for ex vivo data could be obtained. There was no statistically significant difference in the diagnostic sensitivity as well as specificity between ex vivo and in vivo groups.

Table 2

Sensitivity and specificity of HGM for differential diagnosis of nonmelanoma pigmented tumors.