2.1.

Experimental Setup

The laboratory setup is presented in Fig. 1; it combines principles of OCT, RS, and BS for human tissue studies. The setup includes a thermally stabilized semiconductor diode laser module LML-785.0RB-04 ($785\pm 0.1\mathrm{nm}$ central wavelength, 150 mW) for excitation of Raman spectra, a broadband laser diode Broadlighter D-840-HP ($840\pm 50\mathrm{nm}$, 20 mW), and a Michelson interferometer ($50/50$) based on the fiber optic splitter FC850-40-50-APC. The reference arm consists of the collimating lens L1 (CFC-8X-B), focusing lens L2 (AC254-050-B), and adjusted mirror M1 (BB1-E03). An image capture card (PC built-in) digitizes the OCT interference pattern. Test samples were illuminated with excitation radiation via bandpass filtering, which cuts off the Raman shift of the optical-fiber laser. A backscattered spectrum was recorded after sample illumination with a broadband light source (500 to 780 nm). Backscattered radiation and the Raman signal were collected by a fiber probe, which consists of excitation ($100\mu \mathrm{m}$ diameter) and collection fibers ($200\mu \mathrm{m}$ diameter) and a collecting lens L3 (5-mm focal length). The backscattered radiation was passed through a broadband filter to cut off the probing radiation signal. A focusing lens L4 (AC508-250-B) and mirror M2 were used for radiation delivery to receiving channel.

Fig. 1

Experimental setup. L: lens, M: mirror.

Calibration of the system was consequentially performed for OCT, RS, and BS. The overall wavelength error was determined by the produced spectrograph and do not exceed 0.01 nm. As a result, low-noise recording of RS and BS radiations was performed with a spectral resolution of 0.05 nm, using a Shamrock SR-303i spectrograph and an Andor iDus charge-coupled device digital camera. Spatial calibration of the setup was performed with high resolution micrometers and target resolution plate (RES-1). This setup achieved A-scan recordings with a maximum resolution of $3.5\mu \mathrm{m}$.

Raman and backscattered spectra were acquired for a tumorous region of normal skin closely adjacent to a lesion (within 4 cm). Such an approach helps to avoid the influence of the inherent heterogeneity of skin on experimental results. A fiber probe was directly positioned over the tissue sample at a distance of 3 to 5 mm. The beam diameter of the probing radiation was 1 mm. Initially, the recorded BS spectrum was used for dividing the region of interest into healthy and abnormal tissues. The pathological area was subsequently examined by the RS method. The acquisition times were 5 s for BS and 30 s for RS analysis, which allowed for the combination of fast RS scanning with an exact RS examination of select areas on the scanning stage. Three independent measurements were made for each spatial point and then an averaged value was used to take a mean spectrum.

2.2.

Tissue Samples

There were 45 patients (17 female and 28 male, 45 Caucasian, all white, and phenotypes I and II) with skin and lung cancers enrolled in this study. Ex vivo tissue samples were obtained after surgical resection at Samara Regional Clinical Oncology Dispensary under an approved protocol including patient agreement. Tissue samples were stored in sterile boxes at $+4\pm 2°\mathrm{C}$ and were tested using an experimental setup not later than 4 h after resection. All samples were divided into two pieces containing both a healthy tissue region and part of the tumor. One sample piece underwent experimental tests. The rest of the sample was fixed in formalin and prepared for histological analysis. The lung tumor samples were approximately $3\times 3\times 2{\mathrm{cm}}^3$ in size. In total, tests were conducted on 22 ex vivo samples of lung tissue: 11 adenocarcinomas and 11 squamous cell carcinomas. In vivo experiments were performed at the Samara Regional Clinical Oncology Dispensary for 50 skin tissue samples (9 melanomas, 9 basal cell carcinoma, 1 squamous cell carcinoma, 2 pigment nevi, 2 benign tumors, and 27 healthy tissues). The skin tumor samples were approximately $2\times 2\times 1{\mathrm{cm}}^3$ in size. The detailed distribution of human skin and lung lesions, including information about patients and tumor locations, is provided in Table 1. Every tumor study was accompanied by histological analysis to make a final diagnosis. The protocols of in vivo and ex vivo tissue diagnostics were approved by the ethical committee of Samara State Medical University.

Table 1

Summary of patients and lesions.

2.3.

Two-Step Raman Spectroscopy Method

We propose to use a two-step phase-type method for tumor detection and identification in RS studies. In the first step, each measurement is represented as a point on the phase plane of two characteristics, $I_{1320}$ and $I_{1660}$, which are the ratios of maximum scattering intensity in the 1300 to $1340{\mathrm{cm}}^{-1}$ band and the 1640 to $1680{\mathrm{cm}}^{-1}$ band, to the RS intensity in the 1440 to $1460{\mathrm{cm}}^{-1}$ band, respectively. Each data point on the phase plane can be attributed to a specific cancer type or a healthy tissue using histological control as a reference method. On the first step, the lung tumors (adenocarcinoma and squamous cell carcinoma) are separated from healthy tissue, and malignant melanomas-from all other skin cancers, as they are separately allocated in a finite range of the phase plane.

Further differentiation of cancer types is performed in the second step, which is based on analysis of spectral intensity alteration in the tumor tissue in comparison with healthy tissue near the lesion. For this purpose, it was defined as the relative difference (relative index) between RS intensities

2.4.

Data Discrimination Methods

Discriminant analysis (DA) was used to designate tissue classes in the phase planes. DA has the ability to flexibly change priority to favor sensitivity or, on the contrary, to favor specificity. As such, the efficiency of approaches is characterized by their sensitivity and specificity and the ability to select defined classes in different volumes of phase space. The analysis of skin-tumor data allocation was performed using a quadratic discriminant analysis (QDA) classifier, which was determined to be the most effective method for allocation of classes. Hyperquadric curves separate classes in QDA. Their curvature and position depends on the variation of experimental results. The linear discriminant analysis (LDA) has lower levels of specificity, but the sensitivity remains the same. This is due to the fact that LDA successfully separates only similar classes.²⁷

3.1.

Multimodal Tissue Study

Diagnosis of tissue samples was started from OCT image capture followed by BS and RS analysis. Standard OCT image sizes were $15\times 15\times 3{\mathrm{mm}}^3$. The OCT image includes the tumor and healthy tissue areas and, as a result, contains visual information about tumor border configuration, which may be observed by medical specialists.

The next step of the tissue sample analysis was the registration of BS and RS signals, which may be achieved either from a whole sample surface (pixel-by-pixel scanning) or from fixed selected regions. Regions of interest for RS and BS registrations (tumor and healthy points) may be chosen on the basis of OCT image analysis by a medical specialist. Figure 2 represents an example of RS analysis of a melanoma cross section: each point is uniformly located over the cross section and corresponds to the ratio of Raman spectral intensities $I_{1320}/I_{1450}$, captured from an illuminated area with a diameter of 1 mm. As one can see, the spectral analysis of the whole sample is redundant for cancer type identification. It is enough to check the area in the center of the tumor (point 6 or 7 on Fig. 2) and near the boundary in healthy tissue (point 2 or 9 on Fig. 2) for cancer type determination. And boundary itself may be selected by OCT image analysis. Such an approach gives the opportunity to equalize the speed of OCT imaging and RS and BS analysis and simultaneously completes the multimodal examination of the sample.

Fig. 2

Raman spectroscopy (RS) analysis of melanoma for tissues spectral characteristics measurements: points 1, 2, 9, and 10 are healthy tissue, 5, 6, and 7 are tumor, 3, 4, 8—border area.

Thereby, the proposed algorithm of tissue analysis has a number of advantages. It allows one to use multispectral information from different optical methods for tumor diagnosis and reduces the time of tissue study by decreasing the number of analyzing points. Moreover, this approach makes possible step-by-step analysis that altogether can be useful in clinical applications.

3.2.

Optical Coherence Tomography Skin Tissue Study

OCT resolution is 100 to 250 times higher than ultrasound or magnetic resonance imaging, and allows for the identification of very small inhomogeneities in tissue layers. However, it is very difficult to distinguish one type of tumor from another based solely on OCT-image analysis.²⁸ In particular, in our study, only BCC was confidently differentiated by OCT among all examined skin cancer samples. This was due to the presence of a round-like area (a so called “nest”) specific to BCC. Figure 3 shows OCT images of two different types of malignant skin tumors along with their histological images. Melanoma and non-BCC tumors have their own nonspecific topology for each sample. Figure 3(b) shows a melanoma tumor with no clear borders of invasion. One can see only a number of small heterogeneities beneath the upper skin layer.

Fig. 3

Optical coherence tomography (OCT) images of basal cell carcinoma (BCC) (a) and malignant melanoma (MM) (b), $1.5\times 1.5\mathrm{mm}$ each (arrows indicate tumorous areas in healthy skin) and corresponding histologic section of BCC (c) and MM (d), $100\times$ magnification.

It is clearly visible that BCC has a unique rounded topology. However, cases of BCC with a clear nest area were found only in 7 of the 9 BCC samples, and one non-BCC tumor sample also has a nest-like area. The sensitivity for BCC tumor diagnosis by OCT does not exceed 78%, according to the results of this study. Consequently, nest-like topology cannot be taken as the determining criteria for a BCC tumor, and additional spectral analysis must be performed to verify its diagnosis.

The OCT study of skin tissue samples is in agreement with the results from other scientific groups (for example Ref. 29). OCT imaging may provide precise data on 3-D heterogeneities in a sample, but it is difficult to determine a cancer type with high accuracy for non-BBC tumors. Thus, a multimodal approach, with the implementation of BS and RS methods, may improve the accuracy of tumor classification. The simplicity and convenience of joint OCT-RS-BS is based on the possibility of combining radiation sources and recording units. Subsequent filtering of scattered radiation allows an operator to split the OCT signal from the BS and RS signals and process them simultaneously. Then, the spectroscopic data help to identify non-BCC tumors (melanomas and lung carcinomas) and improve sensitivity and specificity in BCC diagnosing.

3.3.

Backscattering Skin Tissue Study

BS analysis allows for the rapid classification of tumors. It is well known that malignant tumors are characterized by a capillary-net expansion and an increase in the concentration of melanin.³⁰ This concentration is proportional to the BS index $M=\lg (R_{650}/R_{700})$, where $R_{\lambda }$ is a BS reflectance coefficient at wavelength $\lambda$. The spatial distribution of capillary inhomogeneity may be estimated by recording the local rate of blood content, which is proportional to the BS index $K=(R_{760}/R_{560})$.

Discriminant analysis was performed for BS data in ex vivo and in vivo experiments.³¹ Averaged BS sensitivity and specificity for ex vivo skin tumor classification were 84% and 90%, respectively. However, for in vivo experiments, the accuracy of BS analysis decreased and did not exceed 83%.

As can be seen from Fig. 4, the $K-M$ phase plane may be used for discriminant analysis of skin tumor types. Here, curve 1 divides the phase plane into two areas: a region of benign tumors (below curve 1) and a region of malignant tumors of different types (above curve 1). Curves 2 and 3 divide the malignant tumor region into several sections, corresponding to tumors of definite types. The maximum values for sensitivity and specificity (86% and 96%, respectively) were obtained for nevus diagnosis. Melanoma diagnosis was characterized by a minimal sensitivity of 82%.

Fig. 4

Discriminant backscattering (BS) analysis of different types of skin tumors.

BS analysis may be used for determining tumor borders based on a combined coefficient $(K-1)+(M-1)/8$, which simultaneously takes into account changes in melanin levels and the concentration of blood in the tumor volume, as well as the individual characteristics of the skin. The distribution of the combined factor on the skin surface allows for the designation of a threshold value (typical for healthy skin), which in turn permits a determination of tumor border area with 1-mm accuracy.

3.4.

Combined Optical Coherence Tomography: Raman Spectroscopy Study of Lung and Skin Tissues

The multimodal OCT and RS unit acquires a high-resolution 3-D image of the examined objects. This allows one to determine the topology of every layer and find an invasion area in healthy tissue. Figure 5 shows an example of these measurements for bronchus tissue. It is easy to find a pathological area in lung and bronchus tissue, since the tumor has clear boundaries. However, we did not find any spatial regularity in the OCT images that would help to differentiate squamous cell carcinomas from adenocarcinomas. This result is in agreement with Ref. 32. Therefore, in lung studies, OCT should be primarily used as an imaging tool for tumor area determination.

Fig. 5

Bronchus sample with marks for RS study: (a) OCT image ($1.5\times 1.5\mathrm{mm}$), (b) digital image ($1.5\times 1.5\mathrm{mm}$); 1—normal tissue, 2—tumor.

The next step is to perform RS evaluation of the tissue in the visualized pathologic area (ignoring the border area). Most known methods of Raman spectral analysis of tumors are based on the definition of threshold criteria for peak intensities in the following Raman bands: 1300 to $1340{\mathrm{cm}}^{-1}$, 1640 to $1680{\mathrm{cm}}^{-1}$, and 1440 to $1460{\mathrm{cm}}^{-1}$. However, the variation in threshold values is quite large. We propose to use the alternative method of two-step Raman analysis described in Sec. 2.3. This method involves patient healthy tissue characteristics, and thus it is more patient-oriented.

For the lung study, such a RS two-step phase analysis is presented in Fig. 6. The intensities $I_{1320}$ and $I_{1660}$ form the phase plane shown in Fig. 6(a), which separates healthy tissue from tumors with high sensitivity and specificity (81% and 84%, respectively). However, the precise determination of tumor type remains nearly impossible (adenocarcinomas and squamous cell carcinomas are mixed in the upper part of the phase plane). In order to resolve different cancer types, the second step in the RS comparison must be used, as shown in Fig. 6(b) for the bands 1450 and $1660{\mathrm{cm}}^{–1}$. Quadratic discriminant analysis was used to separate clusters in the phase plane. Thus, by using the two-step phase analysis, one may increase the accuracy of tumor classification. The two-step method sensitivity and specificity of lung tumor classification are 100% and 82% for adenocarcinoma and 91% and 78% for SCC, respectively.

Fig. 6

Tumor and healthy tissue ex vivo determination for lung and bronchus by two-step phase analysis of Raman spectroscopy (RS) method: (a) first step of RS method, (b) second step of RS method.

In vivo experiments with skin tumors and normal skin tissue were carried out for evaluation of the two-step method accuracy. Figure 7(a) illustrates the first step of the RS method and Fig. 7(b) shows the datasets obtained in the second step of RS quadratic discriminant analysis. In the first step [Fig. 7(a)], melanomas were separated from all other cancer types or healthy skin, with a sensitivity of 78%. The second step [Fig. 7(b)] allows for the identification of additional melanomas that were not detected in the first step. Therefore, overall, a sensitivity of melanoma detection equal to 89% with specificity 88% was achieved. However, the proposed two-step method demonstrates insufficient sensitivity (89%) and specificity (85%) for BCC diagnosis. The two-step method also has an exception for nonpigmented melanoma, which is characterized by a low melanin content and, as a result, produces a weak RS intensity in the 1300 to $1340{\mathrm{cm}}^{–1}$ band.

Fig. 7

In vivo determination of tumorous and healthy tissue in skin by two-step phase analysis of Raman spectroscopy (RS) method: (a) first step of RS method, (b) second step of RS method.