CC BY
2DOI 10.24412/CL-37135-2023-1-27-31
OPTICAL MULTIMODAL STUDY OF SKIN TUMORS
ELINA GENINA1,2, YURY SURKOV1,2, ISABELLA SEREBRYAKOVA1,2, EKATERINA LAZAREVA1,2, YANA KUZINOVA3, SERGEY ZAYTSEV1,4, OLGA KONOPATSKOVA3, DMITRY SAFRONOV3, SERGEY KAPRALOV3, AND VALERY TUCHIN1,2,5
1Optics and Biophotonics Department, Saratov State University, Russia 2Laboratory of Laser Molecular Imaging and Machine Learning, National Research Tomsk State
University, Russia
3Faculty Surgery and Oncology Department, Saratov State Medical University, Russia 4CNRS, CRAN, Université de Lorraine, France 5Institute of Precision Mechanics and Control, FRC "Saratov Scientific Centre of the Russian Academy of Sciences," Russia
ABSTRACT
Despite the development of medicine, cancer remains one of the leading causes of morbidity and mortality worldwide. Most often, if the tumor is diagnosed earlier and treated, the patient will have a better prognosis and much greater opportunities for complete recovery [1,2]. An effective solution to the problem is the use of modern optical technologies, such as light-induced autofluorescence spectroscopy (LIAFS), diffuse reflectance spectroscopy (DRS), Raman spectroscopy, optical coherence tomography (OCT), etc., which are non-invasive methods and provide obtaining diagnostic information in real time. In addition, they are portable and relatively low cost.
LIAFS and imaging relying on endogenous fluorophores represent non-invasive and fast approach for skin cancers diagnosis. However, due to high pigmentation of pigmentary skin nevi and early malignant melanomas of the skin, this method has a significant limitation because of low light penetration depth [3].
DRS is well suited for use in biomedical applications due to its low instrumentation cost, easy implementation. Optical fibers are coupled to a multichannel hyperspectral imaging system, which allows simultaneous acquisition of reflectance spectra from the sample [4].
Neoplastic cells are characterized by increased nuclear material, an increased nuclear-to-cytoplasmic ratio, increased mitotic activity, abnormal chromatin distribution, and decreased differentiation. There is a progressive loss of cell maturation, and proliferation of these undifferentiated cells results in increased metabolic activity. General features of neoplastic cells result in specific changes in nucleic acid, protein, lipid, and carbohydrate quantities and/or conformations. The original analyses for Raman signals are based on differences in intensity, shape, and location of the various Raman bands between normal and cancerous cells and tissues [5]. However, it should be taken into account that high sensitivity to small biochemical changes is accompanied by weak Raman signal often in the presence of high background. Overall, all skin lesions appear to share similar major Raman peaks and bands in fingerprint region. There are no distinctive Raman peaks or bands that can be uniquely assigned to, for example, basal cell carcinoma (BCC) by visual inspection alone. The development of the malignant skin disease increases the content of metabolic products in the pathological areas of the skin, changes the concentration of proteins and lipids [5].
OCT is used for preoperative determination of the peripheral boundaries of BCC in order to choose the optimal treatment method and minimize the invasiveness of surgical intervention. OCT is characterized by high efficiency for the in vivo diagnosis of malignant and benign skin tumors. It can be used for lifetime monitoring of structural changes of the skin during and after treatment of BCC in order to assess its effectiveness. However, due to the limited depth of probing, OCT cannot be used to determine the depth of invasion [6].
High resolution ultrasound (US) imaging systems, which use ultrasound frequencies above 15 MHz can differentiate structures of less than 100 ^m on the beam axis (axial resolution) and 200 ^m on the scan axis (resolution axis). The frequency ranges from 20 to 50 MHz, allowing visualization of the superficial layer of the skin (epidermis and dermis, and the upper part of hypodermis), where the majority of lesions and skin tumors are located. However, the sonographic appearance is often stereotyped: overall anechoic tumor, including large focally dense echoes [7].
In recent years there has been an increasing interest in the combined usage of these technologies that results in increased sensitivity and specificity [2]. Multimodal approaches can increase the effectiveness of early diagnosis and treatment procedures, as well as reduce cancer mortality.
Skin cancers are due to the development of abnormal cells that have the ability to invade or spread to other parts of the body. This is a heterogeneous group of cancers that includes malignant melanoma (MM) and non-melanoma skin cancer, the main ones being basal-cell carcinoma (BCC) and squamous-cell carcinoma (SCC) [8].
MM is the most aggressive form of skin cancer. It is the leading cause of death related to skin disease [1]. The overall mortality rate increases up to 50% in the case of lesions thicker than 4 mm. However, the overall survival of thin lesions (less than 0.5 mm thick) is excellent [2].
BCC is the most common epithelial neoplasm of the skin. It consists of cells similar to the cells of the basal layer of the epidermis and differs from other skin cancers by extremely rare metastasis, but it is capable of extensive local growth, which leads to significant cosmetic and functional disorders [8].
BCC can broadly be divided into three groups, based on the growth patterns. Nodular basal-cell carcinoma includes most of the remaining categories of basal-cell cancer. Nodular BCC (also known as "classic BCC") accounts for 50% of all BCC. It most commonly occurs on the sun-exposed areas of the head and neck. Histopathology shows aggregates of basaloid cells with well-defined borders, showing a peripheral palisading of cells and one or more typical clefts. Central necrosis with eosinophilic, granular features may be also present. The heavy aggregates of mucin determine a cystic structure.5 Superficial BCC is characterized by a superficial proliferation of neoplastic basal-cells. Infiltrative BCC, which also encompasses morpheaform and micronodular BCC, is more difficult to treat with conservative methods, given its tendency to penetrate into deeper layers of the skin. Besides, pigmented BCC exhibits increased melanization [2, 8].
BCC is difficult to distinguish from benign formations by external signs. They can often come in association with other lesions of the skin, such as actinic keratosis, seborrheic keratosis, SCC. Often, patients do not perceive such wounds as skin cancer, but consider them accidental, frivolous "sores" or acne. However, it is prone to extensive local growth, which can lead to its germination deep into the skin and destruction of the surrounding tissue [9, 10].
Thus, multimodal approach in combined with optical clearing increases the effectiveness of methods of optical diagnostics of cancer.
In this study, the development of technologies for biomedical imaging of skin cancer is presented. A combination of high-resolution ultrasound examination and optical methods as Raman spectroscopy, OCT, and DRS were used for differentiation of different BCC subtypes. DRS measurements were combined with the use of biocompatible optical clearing agents to increase the efficacy of MM diagnostics.
Experimental design for basal-cell carcinoma study.
The study involved 40 light-skinned volunteers with BCC and benign neoplasms (BN). Informed consents were acquired from all patients prior to the study. Enrolled patients' age ranged from 40 to 84 years. Among the neoplasms studied were 5 of BN and 26 of BCC, which were divided into the following subtypes: infiltrative-ulcerative (3), pigmented (2), superficial (15), morpheaform (3) and nodular BCC (3). The group of BN included nevi, fibromas and dermatofibromas. At the beginning of the study, a visual examination of the volunteer was carried out by an oncologist. The final diagnosis was made by a specialist based on the results of cytological and/or histopathological studies at the Regional Oncological Clinics No. 2 of Saratov.
The neoplasms were photographed before the study using commercially evaluated Digital Video Dermatoscope DE300 Firefly (tpm taberna pro medicum GmbH, Germany) with 10* zoom and 1920*1080 resolution.
For ultrasonography we used the DUB SkinScanner (tpm taberna pro medicum GmbH, Germany) in B-scan mode with US probes with frequencies of 33 and 75 MHz, scanning depths of 6 and 3.2 mm and longitudinal resolution of 48 and 21 ^m, respectively. The width of the ultrasound scanning window was 13 mm.
To analyze the US data for each neoplasm, 5 US images were selected without artifacts and having the most contrasting borders of the neoplasm with healthy skin. For each ultrasound image, a region of interest (ROI) was marked by the operator, and 12 parameters were calculated depending on the shape and size of the neoplasm: area, perimeter, roundness, eccentricity of the ellipse describing ROI, equivalent diameter of a circle area of a rectangle describing ROI, length of the major axis of the ellipse describing ROI, length of the minor major axis of the ellipse describing ROI, modulus the angle between the major axis of the ellipse describing the ROI and the horizontal axis of the image, the ratio of the area of the ROI to the area of the rectangle describing the ROI, the angle between the maximum diameter of the rectangle describing the ROI and the horizontal axis of the image, the ratio of the perimeter to the area.
Diffuse reflectance was recorded using multichannel reflectance spectrometer USB4000-UV-VIS with a fiberoptic probe QR400-7-VIS-NIR (Ocean Optics, USA) and a spectral range of 450-950 nm. For each neoplasm, at least 5 diffuse reflectance spectra were recorded. To take into account the individual spectral features of the volunteer's skin, for each neoplasm, healthy skin areas nearby or symmetrically located to the lesions were measured.
From the data of the reflectance spectra, the several characteristic coefficients were calculated. The melanin pigmentation index (M) of human skin was determined by the following formula:
m = 100 x (od620 - od7j ,
(1)
where ODX = - lg^x is the effective optical density at the wavelength X.
The erythema index (E) was calculated using the following formula:
E = 100[OD56O + 1.5(OD545 + OD575 ) - 2(OD5lQ + OD^)].
The following expression was used to estimate the index of hemoglobin content (H):
(2)
16
25
The slope coefficient of the diffuse reflectance spectrum (K) in the wavelength range of 650-800 nm and the coefficient reflecting the deviation of the reflectance spectrum from the linear approximation (N) in the wavelength range of 650-800 nm were also calculated.
The coefficient Rt was used to differentiate between malignant melanoma and dysplastic nevi and calculated by the formula:
R R
healthy500 neoplasm700 . ..
Rt = - , (4)
R R
healthy 700 neoplasm 500
where Rhealthy500, Rneoplasm500, Rhealthy700, and Rneoplasm700 are the diffuse reflectance of healthy skin and neoplasms at wavelengths of 500 and 700 nm, respectively.
OCT images were recorded using a GAN930V2-BU (Thorlabs, USA) spectral OCT operating at a central wavelength of 930 nm with an axial resolution of 5.34 ^m and a scanning depth of 2 mm. At least 15 scans were obtained from different areas of the neoplasm.
For texture analysis 180 OCT images without artifacts were selected. _Texture analysis is the analysis of changes in the brightness of pixels within a ROI. This includes the first-order statistical (FOS) methods: mean, standard deviation, skewness, and kurtosis, which are related to the distribution of grayscale pixel intensity and do not depend on interpixel correlation. Second-order statistics (SOS) or gray level coincidence matrices (GLCM), on the contrary, depends on the spatial arrangement of the intensity of the pixels in the ROI. The SOS (or Haralick) parameters include energy, homogeneity, contrast, correlation and entropy.
For each OCT B-scan, ROI containing an image of the internal structure of the neoplasm was chosen. The size of the ROI was not less than 534 ^m in depth and not less than 400 ^m in width.
In order to reduce the effect of noise outside the structural image of the biological tissue, a threshold filter was used. The intensity of each pixel was compared with the threshold value, if the intensity was less than or equal to the threshold value, the intensity of this pixel was taken equal to 0. To determine the threshold value on each OCT scan, a rectangular area approximately 500 by 2000 ^m in size below the biological tissue image was selected, according to which the average and standard deviation of the pixel intensity were calculated. The threshold value was the sum of the mean and standard deviation. To reduce the effect of speckle noise on B-scans, we successively used an erosion filter with a window width of 3 pixels and a 2D smoothing filter with a Gaussian kernel with a standard deviation of 1.5.
To calculate the Haralick parameters, GLCM were calculated for four orientations: horizontal, vertical and two diagonals (directions defined by four angles: 0°, 45°, 90° and 135°). From each GLCM, 5 Haralick characteristics were extracted - energy, homogeneity, contrast, correlation and entropy. Five FOS parameters were calculated from the gray level distribution histogram. In total, 65 characteristic parameters were obtained for each ROI. Pixels with intensity equal to zero, associated with the background, were not taken into account when estimating the FOS and Haralik parameters.
Thus, five observations were obtained for each neoplasm; each observation consisted of 6 DRS parameters, 65 OCT parameters, 12 US parameters. All calculated parameters were compared with the oncologist's diagnosis and combined into a single table, where the columns corresponded to the parameter, the rows to measurements (observations).
To classify the diagnosis and subtype of BCC according to the data of the formed observations, a classifier was built based on the method of k-weighted nearest neighbors. The set number of required nearest neighbors was 10. The Euclidean function was used as the distance metric. To train and evaluate the quality of the model, a cross-validation method was used with data splitting into 5 sections; the number of training iterations was 10. Before the analysis, all parameters were normalized to the maximum value of the parameter among all observations.
Raman spectra were measured with spectrometer QE65000 (Ocean Optics, USA) equipped with 785-nm diode laser and probe (f = 7.5 mm).
Experimental design for melanoma study.
The objects of the study were 9 outbred mice. Transplantation of model skin melanoma was carried out by subcutaneous injection of a suspension of B16F10 melanoma tumor cells into the region of the outer side of the thigh from both sides. The experimental protocol consisted of two stages. The first stage was carried out on day 7 after subcutaneous injection of tumor cell suspension and was performed in vivo. The second stage was carried out ex vivo on day 14.
DRS was used in combination with optical immersion technique using polyethylene glycol-400 (PEG) and aqueous solutions of glucose and sucrose as optical clearing agents (OCA). As enhancers, propylene glycol (PG), oleic acid (OA), and dimethyl sulfoxide (DMSO) were used.
The spectral probe was placed in contact with the central part of the skin areas with subcutaneous melanoma, and the DR spectra of intact skin were registered. Similar DR spectra were also measured near the area of melanoma localization on healthy skin areas. After that, the examined healthy and affected areas of the skin were subjected to dermabrasion for 1 minute using cosmetic device MD3A 933 (Gezatone, France). Further, two areas of mouse skin affected by melanoma, as well as adjacent areas of healthy skin, were treated with therapeutic US for 5 minutes using Dinatron 125 (Dinatronics, USA) in the following mode: CW, 1 MHz, 1Wt/cm2 During treatment, 200 ^l of the OCA + enhancer mixture was applied to the skin under an ultrasound probe. After ultrasonic clarification, the final DR spectra were registered from each area of the skin.
Changes in the slope of the DR spectra in the ranges of 470-520 nm and 650-800 nm after optical clearing were analyzed. Differences in the DR signal in the area of three blood absorption peaks (420 nm, 545 nm and 575 nm) between skin areas with melanoma and healthy skin were estimated using the ratio RbenigJRmelanoma. Results of Multimodal diagnostics of basal-cell carcinoma.
Figure 1 shows the results of evaluations of classification models, a Compatibility Matrix (Validation) based on the nearest neighbor method. Diagonal cells show where observations from the true group were assigned to their corresponding groups. It is clearly seen that the combination of DRS, OCT and US leads to an increase in true positive rates (TPR). TPR has maximal values for surface BCC (100%) and minimal for pigmented BCC (60%)_
Method
Number of observations
Predicted class
TPR/FNR
m Pi Q
Infiltrative-ulcerative BCC
Pigmented BCC Surface BCC Morpheaform BCC Nodular BCC
37 1 9 3
3 5 1 6
2 1 4 3
74 1
1 3 11
4 13
Infiltrative-ulcerative BCC
Pigmented BCC Surface BCC Morpheaform BCC Nodular BCC
T
U О
Predicted class
Infiltrative-ulcerative BCC
Pigmented BCC Surface BCC Morpheaform BCC Nodular BCC
42 1 4 1 2
11 4
3 1 3 2 1
4 2 66 3
2 1 4 8
1 8 1 5
BN M.iW 2,0« 8,0% 2,0% 4,0%
Infiltrative-ulcerative BCC 73, 3% 26,7%
Pigmented BCC 30,0% 10,0% 30,0% 20,0% 10,0%
Surface BCC 3,3% 2,7% sa,o% 4,0%
Morpheaform BCC 1 :„ 6,7% 26,7% 53.35S
Nodular BCC 6,7% 53,3% 6.7% 33,3%
84,0% 16,0%
73,3% 26,7%
30,0% 70,0%
38,0% 12,0%
53,3% 46,7Я
33,3% 66,7%
CO
Predicted class
Infiltrative-ulcerative BCC
Pigmented BCC Surface BCC Morpheaform BCC Nodular BCC
45 3 1 1
11 2 2
2 8
2 73
3 9 2 1
4 5 2 4
Infiltrative-ulcerative BCC
Pigmented BCC Surface BCC Morpheaform BCC Nodular BCC
£
«в
T
о о «в m R
Q
Infiltrative-ulcerative BCC
Pigmented BCC Surface BCC Morpheaform BCC Nodular BCC
49 1
12 3
1 6 3
75
1 12 2
2 13
Infiltrative-ulcerative BCC
Pigmented BCC Surface BCC Morpheaform BCC Nodular BCC
Predicted class
BN
BN
BN
BN
BN
100%
BN
BN
Figure 1: The indicators of validation of the classification oof neoplasms, true positive rates (TPR) are marked in blue, false negative rates (FNR) are marked in orange. Rows show the true class, columns show the predicted class. The values in the empty cells correspond to 0.
Raman spectra for normal skin tissue and of nodular BCC before and after treatment are depicted in Figure 2. Proteins predominantly contribute to the appearance of bands in the spectral range 1240-1270, 1340, 1440-1460, and 1665 cm-1, the spectral features arising from the contribution of lipids, are observed in the 1271-1301, 1440,1650-1660 cm 1 bands.
Figure 2: Raman spectra of normal skin tissue, BN, and nodular BCC before and after treatment
One of the significant differences between malignant and benign formations is the process of metabolism and destruction of collagen. Cells of malignant tumors form fast-growing, low-differentiated structures, and the development of such structures is accompanied by the increased activity of collagenase. Collagenase destroys the molecular bonds of collagen fibers, and changes in Raman spectra of skin tissue can be observed in 1248, 1454, and 1665 cm-1 bands associated with peaks of collagen [5]. In the BCC, there is an increased content of proteins (430, 475 cm-1) and nucleic acids (622, 685 cm-1), a decreased content of lipids (1287, 1419 cm-1) and keratin (1463, 1670 cm-1). Increased peaks associated with DNA (755 cm-1) and cell nuclei (831 cm-1).
Results of Multimodal diagnostics of melanoma. All mixtures of OCA to varying degrees after enlightenment showed an increase in differences in the DR signal from the three peaks of blood absorption between the skin areas with melanoma and healthy skin. The greatest differences after skin optical clearing in vivo were demonstrated by a mixture of PEG/OA/PG (an increase of ~2 times after the clearing relative to intact skin at the wavelength 420 nm (Soret absorption band).
The reduction in skin scattering, and, as a result, the best effect of optical clearing, based on the analysis of relative changes in the Rbenign/Rmelanoma ratio, was achieved at the ex vivo stage of the experiment with a mixture of Glucose/OA/PG. At the wavelengths 420 nm, 545 nm, and 575 nm, the relative changes in the Rbeni^/Rmelanoma ratio after clearing were 314%, 205%, and 204%, respectively.
Thus, the sizes of neoplasms were evaluated using ultrasound examination, and their internal structure was visualized using OCT in combination with OC. Diffuse reflectance can extract physiological parameters such as hemoglobin content, oxygen saturation, and tissue microarchitecture. Raman spectroscopy is helpful for determining lipid, nuclear, and protein content.
Our results demonstrate the ability of these modalities to quantitatively assess tissue biochemical, structural, and physiological parameters that can be used to determine tissue pathology. Our future work will be extended on the use of all presented modalities, including skin optical clearing. The study was funded by RFBR grant number 20-52-56005. REFERENCES
[1] World Health Organization. Cancer. https://www.who.int/health-topics/cancer#tab=tab_3
L. Lim, B. Nichols, M. R. Migden, N. Rajaram, J. S. Reichenberg, M. K. Markey, M. I. Ross, J.W. Tunnell, Clinical study of noninvasive in vivo melanoma and nonmelanoma skin cancers using multimodal spectral diagnosis, J. Biomed. Opt. 117003-117003, 2014.
E. Borisova, E. Carstea, L. Cristescu, E. Pavlova, N. Hadjiolov, P. Troyanova, L. Avramov, Light-induced fluorescence spectroscopy and optical coherence tomography of basal cell carcinoma, J. Innovat. Opt. Health Sci. 261-268, 2009.
E. Borisova, E. Nikolova, P. Troyanova, L. Avramov, Autofluorescence and diffuse reflectance spectroscopy of pigment disorders in human skin, J. Optoelectron. Adv. Materials 717-722, 2008.
Multimodal optical diagnostics of cancer, V. V. Tuchin, J. Popp, and V. Zakharov, (Eds.), Springer Nature, Cham, Switzerland p. 3-597, 2020.
W. Gao, V.P. Zakharov, O.O. Myakinin, I.A. Bratchenko, D.N. Artemyev, D.V. Kornilin, Medical images classification for skin cancer using quantitative image features with optical coherence tomography, J. Innovat. Opt. Health Sci. 1650003, 2016.
M. Laurent, M. Samimi, G. Georgesco, Y. Mourtada, M. Naouri, J.M. Grégoire, F. Ossant, F. Patat, L. Vaillant, High resolution ultrasound imaging of melanocytic and other pigmented lesions of the skin, Chapter 8, Ultrasound imaging (Ed. M. Tanabe), InTech, London, UK, p. 139-150, 2011.
G. Paolino, M. Donati, D. Didona, S. R. Mercuri, C. Cantisani, Histology of non-melanoma skin cancers: an update, Biomedicines, 71, 2017.
N. Fusco, G. Lopez, U. Gianelli, Basal-cell carcinoma and seborrheic keratosis: When opposites attract, Int. J. Surg. Pathol. 464, 2015.
[10] W. James, D.M. Elston, J.R. Treat, M.A. Rosenbach, I.M. Neuhaus, Andrews' diseases of the skin: Clinical dermatology, 13th Edition, Elsevier, North York, Canada, p. 1008, 2020.
[2]
[3]
[4]
[5]
[6]
[7]
[8] [9]
