Diagnostic performances of skin cancers using intensity- and gradient-based features with optical coherence tomography: a pilot study

Introduction

Skin cancer is one of the most prevalent cancers in the world, with more than 1.5 million cases in 2020 worldwide (1). In the United States, skin cancers claim lives, with 8,430 deaths in 2025, representing nearly 20 deaths per day (2). Among skin cancers, amelanotic melanoma (AM) is a subtype of cutaneous melanoma with little or no pigment on visual or histopathologic examination and accounts for approximately 2% of melanoma cases (3). AM is often misdiagnosed or diagnosed in late stages due to its variable clinical presentation. To benefit from advances in early detection and treatment, the skin biopsy, as the gold standard, is used for diagnosing skin cancers in clinic. However, the skin biopsy is a time-consuming and invasive method that may cause high levels of discomfort to the patients (4).

Several technologies have been developed as novel non-invasive diagnostic methods, including dermoscopy, reflectance confocal microscopy (RCM), Raman spectroscopy (RS), and optical coherence tomography (OCT) (5). Dermoscopy is a non-invasive in vivo technique that collects the radiation in the visual range from upper skin layers which allows the visual inspection of microstructures of the epidermis, the dermal-epidermal junction (DEJ), and papillary dermis (6). The most important disadvantage is that dermoscopy needs to be performed by experienced examiners since dermoscopy by untrained or less experienced examiners was found to be no better than clinical inspection without dermoscopy (7). RCM generates the image contrast based on the natural differences in refractive indices of subcellular structures within the tissues (8). Although the maximum imaging depth is about 300 µm, the imaging resolution of RCM decreases substantially below a depth of 100 to 150 µm, restricting accurate diagnostic interpretation to the epidermis and superficial dermis (9). RS detects changes in the chemical composition of skin by comparing the Raman scattering wavelength shifts to a molecular fingerprint (10). However, RS requires to acquire sufficient spectra per skin sample to accommodate different cells’ variations, while consensus approaches are lacking in sampling size evaluation required to capture the targeted level of diversity and complexity in environmental systems (11). OCT uses scanning low-coherence interferometry to produce depth-resolved images of optical backscatter originating from local discontinuities of the refractive index within tissue microstructures (12). As a non-invasive optical modality, OCT can generate the cross-sectional images in deep regions of biological tissue.

The first experimental OCT device was implemented to image the human retina in vivo in 1991 (13). In 1997, OCT was first described as a potential imaging method for dermatology (14). For example, OCT was able to observe burn wounds in a noninvasive manner, thereby reducing the pain and discomfort associated with invasive procedures (15,16). The evaluation of skin tissues with OCT has established an important research direction that allows the observation of the morphological changes in skin tissues. Particularly, various features were extracted from OCT images to assist in the diagnosis of skin cancers. Chen et al. used the nodular features of the tumor, mucin surrounding the tumor, tumor subtype, and necrosis to diagnose the basal cell carcinoma (BCC) (17). Mtimet et al. used the features of lobules, millefeuille pattern, clefting, bright rim, lobules connected to epidermis, hemispheric lobules, flat and disrupted DEJ, atrophic epidermis, bright and stretched stroma, perilobular, dilated, branching and compressed vessels from OCT images to differentiate BCC and non-BCC (18,19). Cinotti et al. characterized squamous cell carcinoma (SCC) in OCT by a thickened epidermal layer, which may hamper the visualization of the underlying structures, and a loss of clear demarcation of the DEJ (20). Welzel et al. found that melanomas contained more blood vessel features in irregular vascular shapes such as blobs, coils, curves and serpiginous vessels (21). The morphological features of skin cancers can be captured in OCT images because the pixels in morphological features in OCT images have the similar intensities. Therefore, the intensity-based parameters could be used in the diagnosis of skin cancers. Marvdashti et al. used the intensity-based features and textural features to detect BCC (22). Adabi et al. used the textural features and other higher-order statistical features to differentiate BCC and SCC from healthy tissue (23). Bedin et al. used the higher-order statistical feature—fractal dimension (FD) as the prognostic factor to describe nuclear chromatin measured in routine histological preparations of malignant melanomas (24). Our previous study showed that the intensity- and gradient-based features extracted from OCT images could be used to differentiate the melanomas from pigment nevi (25). Generally speaking, the intensity-based features, textural features and FD are the first-, second- and higher-order statistical features. The second-order statistical features such as the textural features and the higher-order statistical features such as the FD might be treated as the intensity-processed parameters, which are time-consuming compared to the first-order statistical feature—intensity-based features. However, the diagnostic power of these intensity- and gradient-based features in skin cancers, including AM, has not been determined. Therefore, in this study, the diagnostic performances of intensity- and gradient-based features will be investigated for diagnosing skin cancers with OCT. We present this article in accordance with the STARD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2438/rc).

Methods

OCT system

The spectral domain OCT (SD–OCT) system was built in the laboratory in the Department of Laser and Biotechnical Systems at Samara University. The system’s schematic is shown in Figure 1 (25). The core component of this custom-built device is the Michelson interferometer, which is able to split the incident light at a 50/50 ratio along the sample and the reference arms. The system includes a diffraction grating that can provide 1,200 grooves per millimeter and a charge-coupled device (CCD) line scan camera that can image with a resolution of 4,096 pixels and a 29.3 kHz line rate. The type of image card used in the device for the digitization of the signal is NI-IMAQ PCI-1428. The device’s output power is 14 mW, with the central wavelength of the light source being 840 nm. The axial and lateral resolutions of the OCT system are equal and approximately 6 µm.

Figure 1 Custom-built SD–OCT system. 1: broadband light source; 2: 50/50 beam splitter; 3: sample arm; 4: reference arm; 5: spectrometer with grating; 6: charge-coupled device (CCD) camera; 7: computer with IMAQ. This figure was adapted from an Open Access article (25) under the terms of the Creative Commons CC BY license. SD–OCT, spectral domain optical coherence tomography.

Data collections

This prospective study is the outcome of an international collaboration. The participants were recruited at Samara State Medical University (Samara, Russian Federation). The OCT images were collected at Samara University (Samara, Russian Federation). Image processing and analysis were performed at Ningbo University of Technology (Ningbo, China) and The First Affiliated Hospital of Ningbo University (Ningbo, China). The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the ethical committee of Samara State Medical University (Samara Region, Samara, Russia, protocol No. 132, 29 May 2013). Informed consent was taken from all the participants.

In this study, four types of skin tissues were used as follows: (I) superficial BCC, a common form of skin cancer; (II) AM, an uncommon skin cancer that is often misdiagnosed or diagnosed in late stages; (III) pigmented nevus tissues most of which are benign but could potentially lead to skin cancer; and (IV) normal skin tissue that is used as the control group. From May 2017 to December 2019, 5 patients with non-pigmented melanomas, 8 patients with BCC, 4 subjects with pigment nevi, and 3 normal subjects were randomly recruited in the department of dermatology at Samara State Medical University (Samara, Russian Federation). Inclusion criteria were: (I) age ≥18 years; (II) presence of a skin lesion clinically indicated for biopsy; and (III) willingness to provide written informed consent. Exclusion criteria were: (I) lesions in anatomically difficult locations for OCT imaging; or (II) a history of prior treatment to the target lesion. For each enrolled patient, a study clinician documented the lesion’s anatomic location, clinical appearance, and preliminary clinical diagnosis. The skin tissues were first imaged using OCT and then biopsied to determine the categories that were available to researchers. Each sample was scanned in 500 cross-sectional OCT images by using the custom-built SD-OCT device. Five OCT images for each subject were randomly selected from the 500 OCT images. The size of a raw image was 500×2,048 pixels. However, pixels in deep depth of the raw OCT image did not represent the true biological tissue. Thus, the images in this study were reduced to 500×600 pixels by removing the meaningless pixels starting from the 601st pixel in the axial direction.

OCT image processing

The intensities in the OCT image may not accurately represent the reflected light since intensities may mix with noise. The typical noise type is “speckle” and is attributed to the physics of light-tissue interaction. Therefore, the identification and elimination of “speckle” noise and other noise types from OCT images is the first processing step that needs to be performed in OCT image processing. First, the intensities in OCT images were normalized in the range of [0, 1]. Second, the background noise was eliminated from all OCT images. The background noise is any light other than the light being monitored. The background noise types include environmental light, such as sun rays, and illumination light in laboratories. The reduction of the background noise is important in the field of image noise removal. Thus, we used a simple methodology to remove the background noise. We selected a small region with a fixed size in each OCT image at the top edge of each OCT image. The value of the average intensity in this region was then calculated and regarded as background noise. The noise-filtered OCT image was then obtained by subtracting the background noise from the original OCT image. Third, the speckle noise was removed from each OCT image. The speckle noise formed a grainy spot pattern in OCT images that obscured low-intensity regions and small features, caused blurring, and altered the OCT image contrast. Thus, it became impossible to obtain the accurate intensity-based features from the OCT images. To remove the speckle noise from OCT images, the interval type II fuzzy anisotropic diffusion filter was employed because it could effectively improve signal-to-noise by combining the anisotropic diffusion filter and interval type-II fuzzy system (26).

Features

The OCT images were formed by the reflected light from biological tissue. The reflectance in OCT images contained the information about the optical properties of the biological tissue, which the main sources of reflection are collagen fiber bundles and the dark areas in the images represent the homogeneous material with low reflectivity. Therefore, the reflectivity and its distributions in OCT images may represent the structural and optical feature changes and might be used as the indicator to classify the diseased tissue. An image gradient is a directional change in the intensity of an image. Image gradient may be used to extract information such as edge or shape from images. In this study, 10 intensity- and gradient-based features were employed. The mean, median, variance, skewness and kurtosis of intensities and gradients were extracted from OCT images. The equations of intensity- and gradient-based features are listed in Table 1.

Additionally, texture features, characterizing the complexity and pattern of structure in the skin tissue, have been suggested for use in the diagnosis of skin cancers using OCT (23). Therefore, texture features, including contrast, correlation, energy, entropy, and homogeneity, were calculated from OCT images, and were used for comparing with intensity- and gradient-based features. The equations of texture features are listed in Table 2.

In addition to intensity-based features, the mean, median, variance, kurtosis, and skewness of gradients were also extracted from OCT images, in which the intensity I in equations of Table 1 was replaced by the gradient:

G=|∂I∂xi→+∂I∂yj→|[1]

Statistical analysis

One-way multivariate analysis of variance (MANOVA) followed by Tukey post-hoc test was performed to compare the intensity- and gradient-based features, and texture features between study groups using SPSS Statistics v.27 (IBM Corp, Armonk, NY, USA). Variables showed statistically significant difference between study groups when P<0.05. The receiver operating characteristic (ROC) analysis was performed to determine the diagnostic performances of parameters by using SPSS Statistics v.27 (IBM Corp). An area under the ROC (AUROC) curve was calculated.

Results

The OCT images of superficial BCC and AM are shown in Figure 2. The OCT image of superficial BCC (Figure 2A) has easily observable flat, scaly with indistinct borders, which is consistent with its histologic characters (27). The OCT image of AM (Figure 2B) shows diffusely scattering cells are located under the epidermis layer, while the normal epidermis is seen as a bright stripe on the tissue surface, validating by its histology (3).

Figure 2 Optical coherence tomography images of basal cell carcinoma (A) and amelanotic melanomas (B).

Results of the intensity- and gradient-based features, and texture features are shown in Tables 3,4. All data are expressed as the mean ± standard deviation. Table 5 and Table 6 show intensity- and gradient-based features and texture features with the diagnostic performances (AUROC >0.7, sensitivity >0.7, specificity >0.7) between study groups, respectively.

Diagnostic performances of intensity- and gradient-based features

The variance of intensity, the mean and median of gradient demonstrated high diagnostic performances for distinguishing nevi from normal skin tissue. The highest accuracy was achieved by using the variance of intensity (AUROC =0.920, sensitivity =1.000, specificity =0.750). In differentiating BCC from normal skin tissue, the variance of intensity, the mean and median of gradient demonstrated high diagnostic performances. The highest accuracy was achieved by using the mean of gradient (AUROC =0.932, sensitivity =1.000, specificity =0.775). The variance of gradient showed high diagnostic performance (AUROC =0.719, sensitivity =0.700, specificity =0.875) to differentiate BCC from nevi. The variance, skewness, and kurtosis of gradient demonstrated high diagnostic performances for distinguishing AM from nevi. The highest accuracy was achieved by using the skewness of gradient (AUROC =0.828, sensitivity =0.750, specificity =1.000). Moreover, in differentiating AM from BCC, the skewness and kurtosis of intensity and gradient demonstrated high diagnostic performances. The highest accuracy was achieved by using the skewness and kurtosis of intensity (AUROC =0.900, sensitivity =0.875, specificity =1.000).

Diagnostic performances of texture features

The correlation and homogeneity demonstrated high diagnostic performances for distinguishing nevi from normal skin tissue. The highest accuracy was achieved by using the correlation (AUROC =0.933, sensitivity =1.000, specificity =0.800). The correlation, energy, and homogeneity showed high diagnostic performances to differentiate BCC from normal skin tissue. The highest accuracy was achieved by using the correlation (AUROC =0.983, sensitivity =1.000, specificity =0.875). And the correlation, energy, and homogeneity also showed high diagnostic performances in differentiating AM from normal skin tissue. The highest accuracy was achieved by using the homogeneity (AUROC =0.869, sensitivity =1.000, specificity =0.800).

Discussion

In this study, the diagnostic performances of intensity- and gradient-based features, and texture features for distinguishing the skin cancers using OCT were investigated. It was found that the intensity- and gradient-based features showed the diagnostic power to classify the different skin tissues. Particularly, the skewness and kurtosis of intensity and gradient were capable of distinguishing AM from BCC with high accuracy. Meanwhile, texture features, including correlation, energy, and homogeneity, also demonstrated robust abilities to differentiate the skin cancers from normal skin tissues.

Compared to nevi, the significantly lower variance of gradient in AM and BCC suggests that the changes of the reflected light in malignant lesions are more stable than in benign ones. This stability aligns with the known loss of architectural complexity in malignancies (28), where disorganized collagen and cellular atypia paradoxically yield smoother intensity gradient transitions in OCT signal. The smoothness of gradients in AM and BCC further reflects a pathological simplification, where chaotic biology manifests as deceptively ordered optical patterns. The enlarged epithelioid cells in AM and the basaloid cell nests in BCC exhibit uniform optical scattering properties, suppressing local intensity fluctuations that normally arise from heterogeneous tissue interfaces, and leading to the smoother, more uniform OCT signal gradient profiles. Therefore, the variance of gradient from OCT images could serve as a quantitative biomarker for malignancy.

And it was found that the skewness and kurtosis of intensity and gradient distributions in AM were significantly different from those in BCC. The statistical results indicated that BCC may generate more asymmetric and sharper peaks in intensity and gradient distributions than AM. It suggests that the structural heterogeneity within BCC lesions generates subtle but detectable higher-order statistical imbalances in the OCT signal due to its nested architecture and abrupt DEJ disruption. While the more homogeneous cellularity and gradual interface transitions in AM yield more Gaussian-like distributions, which reflect its biologically indolent growth pattern and less pronounced optical scattering heterogeneity. It is consistent with the clinical findings that AM often evades early detection not due to signal complexity, but because of its diffuse homogeneous signal in images (29,30). Therefore, this statistical finding offers a quantifiable biomarker for differentiating these two challenging tumors in vivo.

Although co-occurrence matrix-derived texture features have been used to assist the diagnosis of various diseases with OCT and they showed the abilities to differentiate AM, BCC, and nevi from normal skin tissue, they did not have the ability to distinguish AM from BCC. In contrast, our intensity- and gradient-based features achieve up to 87.5% sensitivity and 100% specificity in distinguishing AM from BCC. It suggests that these intensity- and gradient-based features are not only the statistical numbers but also optical features which can capture subtle interface irregularities and optical homogeneity when confronting tumors with overlapping structural subtleties. Compared to texture features, our approach provides a more accessible method for classifying skin cancers with OCT in clinic.

There are some limitations in this study. First, the limited sample sizes might restrict the application of this methodology in classifying skin cancers. Second, the quality of images using our custom-built SD-OCT system. Third, the reproducibility of the findings needs to be determined. In the future, it could be improved by employing more subjects, upgrading our custom-built SD-OCT system to obtain high quality of images, and validating our findings with more OCT devices. Nevertheless, our study is a pilot study, which has already demonstrated that the intensity- and gradient-based features had the ability to classify the skin cancers using OCT in clinic.

Conclusions

To classify the skin cancers in a non-invasive manner using OCT, the intensity- and gradient-based features from OCT images were employed. The diagnostic performances of intensity- and gradient-based features with texture features were evaluated. Analyses of variance and ROC were performed on four types of skin tissues, including the AM, BCC, pigment nevi, and normal skin tissue. The skewness and kurtosis of intensity could achieve high diagnostic performance in differentiating AM from BCC (AUROC =0.900, sensitivity =0.875, specificity =1.000). Our findings suggest that intensity- and gradient-based features could capture subtle interface irregularities and optical homogeneity for classifying skin cancers.

Acknowledgments

We gratefully acknowledge the data and technique support provided by the Samara State Medical University, Samara University, The First Affiliated Hospital of Ningbo University and sincerely appreciate the valuable contributions and dedicated efforts of all colleagues and staff members involved in this study.

Footnote

Reporting Checklist: The authors have completed the STARD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2438/rc

Data Sharing Statement: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2438/dss

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2438/prf

Funding: This research was supported by Natural Science Foundation of Ningbo (No. 2024J411), the Health Major Science and Technology Planning Project of Zhejiang Province, China (No. WKJ-ZJ-2411), China Medical Education Association–“Hengjing” Skin Immunology Special Research Project (No. HJ-DER-2024-43), Ningbo Public Welfare Research Program (No. 2023S180), and by the Ministry of Education and Science of the Russian Federation.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2438/coif). L.L. reports receiving support from Natural Science Foundation of Ningbo. B.L. reports receiving support from China Medical Education Association–“Hengjing” Skin Immunology Special Research Project. S.X. reports receiving supports from the Health Major Science and Technology Planning Project of Zhejiang Province, China and the Ningbo Major Research and Development Plan Project. I.A.B. reports receiving support from the Ministry of Education and Science of the Russian Federation. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the ethical committee of Samara State Medical University (Samara Region, Samara, Russia, protocol No. 132, 29 May 2013). Informed consent was taken from all the participants.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Jagemann I, Wensing O, Stegemann M, et al. Acceptance of Medical Artificial Intelligence in Skin Cancer Screening: Choice-Based Conjoint Survey. JMIR Form Res 2024;8:e46402. [Crossref] [PubMed]
2. The Global Statistics. Skin Cancer Statistics in the U.S. 2025. 2025. Available online: https://www.theglobalstatistics.com/united-states-skin-cancer-statistics/
3. Jiang C, Jain NP, Stewart CL. Amelanotic melanoma: Clinical presentation, diagnosis, and management. Clin Dermatol 2025;43:10-5. [Crossref] [PubMed]
4. Marcinkevics Z, Aglinska A, Rubins U, et al. Remote Photoplethysmography for Evaluation of Cutaneous Sensory Nerve Fiber Function. Sensors (Basel) 2021;21:1272. [Crossref] [PubMed]
5. Cantisani C, Baja AV, Gargano L, et al. Optical Coherence Tomography as a Valuable Tool for the Evaluation of Cutaneous Kaposi Sarcoma Treated with Imiquimod 5% Cream. Diagnostics (Basel) 2023;13:2901. [Crossref] [PubMed]
6. Sgouros D, Routsi E, Evangelodimou A, et al. Use of Dermoscopy among Greek Dermatologists in Everyday Clinical Practice: A National Questionnaire-Based Study. J Clin Med 2024;13:972. [Crossref] [PubMed]
7. Lin RZ, Amith MT, Wang CX, et al. Dermoscopy Differential Diagnosis Explorer (D3X) Ontology to Aggregate and Link Dermoscopic Patterns to Differential Diagnoses: Development and Usability Study. JMIR Med Inform 2024;12:e49613. [Crossref] [PubMed]
8. Krishnamurthy S, Brown JQ, Iftimia N, et al. Ex Vivo Microscopy: A Promising Next-Generation Digital Microscopy Tool for Surgical Pathology Practice. Arch Pathol Lab Med 2019;143:1058-68. [Crossref] [PubMed]
9. Levine A, Markowitz O. Introduction to reflectance confocal microscopy and its use in clinical practice. JAAD Case Rep 2018;4:1014-23. [Crossref] [PubMed]
10. Ye H. Burn-related Collagen Conformational Changes in ex vivo Porcine Skin using Raman Spectroscopy. Sci Rep 2019;9:19138. [Crossref] [PubMed]
11. Wu M, Gao B, Wei X. Recent advances in Raman spectroscopy for skin diagnosis. J Innov Opt Health Sci 2023;16:2330003. [Crossref]
12. Burri C, Salzmann S, Wandel J, et al. Real-time OCT feedback-controlled RPE photodisruption in ex vivo porcine eyes using 8 microsecond laser pulses. Biomed Opt Express 2023;14:6328-49. [Crossref] [PubMed]
13. Huang D, Swanson EA, Lin CP, et al. Optical coherence tomography. Science 1991;254:1178-81. [Crossref] [PubMed]
14. Welzel J, Lankenau E, Birngruber R, et al. Optical coherence tomography of the human skin. J Am Acad Dermatol 1997;37:958-63. [Crossref] [PubMed]
15. Li H, Bu Q, Shi X, et al. Non-invasive medical imaging technology for the diagnosis of burn depth. Int Wound J 2024;21:e14681. [Crossref] [PubMed]
16. Lindert J, Tafazzoli-Lari K, Tüshaus L, et al. Optical coherence tomography provides an optical biopsy of burn wounds in children-a pilot study. J Biomed Opt 2018;23:1-6. Erratum in: J Biomed Opt 2019;24:1. [Crossref] [PubMed]
17. Chen S, Xie F, Hao T, et al. Evaluation of ultrahigh-resolution optical coherence tomography for basal cell carcinoma, seborrheic keratosis, and nevus. Skin Res Technol 2021;27:479-85. [Crossref] [PubMed]
18. Mtimet L, Boussingault L, Aktas D, et al. Line-field confocal optical coherence tomography of basal cell carcinoma: A retrospective study on diagnostic performance. J Eur Acad Dermatol Venereol 2025;39:1468-80. [Crossref] [PubMed]
19. Adan F, Mosterd K, Kelleners-Smeets NWJ, et al. Diagnostic Value of Optical Coherence Tomography Image Features for Diagnosis of Basal Cell Carcinoma. Acta Derm Venereol 2021;101:adv00607. [Crossref] [PubMed]
20. Cinotti E, Tognetti L, Cartocci A, et al. Line-field confocal optical coherence tomography for actinic keratosis and squamous cell carcinoma: a descriptive study. Clin Exp Dermatol 2021;46:1530-41. [Crossref] [PubMed]
21. Welzel J, Schuh S, De Carvalho N, et al. Dynamic optical coherence tomography shows characteristic alterations of blood vessels in malignant melanoma. J Eur Acad Dermatol Venereol 2021;35:1087-93. [Crossref] [PubMed]
22. Marvdashti T, Duan L, Aasi SZ, et al. Classification of basal cell carcinoma in human skin using machine learning and quantitative features captured by polarization sensitive optical coherence tomography. Biomed Opt Express 2016;7:3721-35. [Crossref] [PubMed]
23. Adabi S, Hosseinzadeh M, Noei S, et al. Universal in vivo Textural Model for Human Skin based on Optical Coherence Tomograms. Sci Rep 2017;7:17912. [Crossref] [PubMed]
24. Bedin V, Adam RL, de Sá BC, et al. Fractal dimension of chromatin is an independent prognostic factor for survival in melanoma. BMC Cancer 2010;10:260. [Crossref] [PubMed]
25. Gao W, Zakharov VP, Myakinin OO, et al. Medical images classification for skin cancer using quantitative image features with optical coherence tomography. J Innov Opt Health Sci 2016;9:1650003. [Crossref]
26. Puvanathasan P, Bizheva K. Interval type-II fuzzy anisotropic diffusion algorithm for speckle noise reduction in optical coherence tomography images. Opt Express 2009;17:733-46. [Crossref] [PubMed]
27. Basset-Seguin N, Herms F. Update in the Management of Basal Cell Carcinoma. Acta Derm Venereol 2020;100:adv00140. [Crossref] [PubMed]
28. Pradhan S, Clary JM, Seliktar D, et al. A three-dimensional spheroidal cancer model based on PEG-fibrinogen hydrogel microspheres. Biomaterials 2017;115:141-54. [Crossref] [PubMed]
29. Gunbey HP, Gunbey E, Sayit AT, et al. Magnetic resonance imaging (MRI) appearances of primary amelanotic malign melanoma in the nasal cavity: a rare case. J Clin Diagn Res 2015;9:TD01-3. [Crossref] [PubMed]
30. Kim SS, Han MH, Kim JE, et al. Malignant melanoma of the sinonasal cavity: explanation of magnetic resonance signal intensities with histopathologic characteristics. Am J Otolaryngol 2000;21:366-78. [Crossref] [PubMed]