Use of Pixel Intensity Measurements Derived from OCT Images to Differentiate Between Seborrheic Keratosis and Melanomas: A Pilot Study

Abstract

Differentiating Seborrheic Keratosis (SK) from melanoma can be difficult based on visual observations and dermoscopy since both are pigmented lesions. While SK is considered a benign lesion that is localized, in contrast melanoma can spread to other tissues and lead to death if it metastasizes. Therefore, it is important to be able to noninvasively differentiate between SK and melanoma to limit the number of unnecessary biopsies performed. We have measured the pixel intensity of Optical Coherence Tomography (OCT) images of normal skin, SK, and melanoma by breaking OCT images into low (green), medium (blue) and high (red) pixel intensity vs. depth images. Normal skin and SK are characterized by higher green scale pixel intensity vs. depth plots while melanoma has a lower green scale pixel intensity vs. depth plot. Melanoma also has lower red scale pixel intensity vs. depth plot compared to SK and normal skin. Our results show that a decreased pixel intensity of the superficial epidermis that is observed in melanomas is likely due to formation of melanin aggregates that approach the wavelength of light in size. The decreased pixel intensity of melanoma is likely a result of increased amounts of melanin particles in melanocytes and keratinocytes. The specificity and sensitivity of differentiating SK and melanoma and normal skin from melanoma based on quantitative pixel intensity vs. depth are about 85% to 100%, respectively. The sensitivity and specificity of differentiating normal skin from SK is 60% and 100%, respectively. These results suggest that color-coded OCT images can be used to noninvasively screen for melanomas along with dermoscopy and visual inspection. The ability to collect OCT lesion data noninvasively in concert with remote data acquisition will allow rapid patient screening for melanomas in areas where dermatologist visits are difficult to schedule.

Introduction

Cutaneous Melanoma (CM) is a malignancy arising particularly in white populations [1]. It is related to UV light exposure from direct exposure to sunlight and indoor tanning beds. Changes in sunbathing habits leading to more ultraviolet radiation exposure and an increasing use of indoor tanning beds play a central role in the rise of CM that has been reported from all Caucasian populations studied [1-3]. CM is typically a highly visible cancer due to the presence of varying amounts of brown pigment. The incidence of CM has been rising rapidly in many populations and melanoma survival is correlated with tumor thickness [1-3]. Therefore, early detection is a strategy for reducing the incidence of malignant melanoma. CM is a particularly aggressive form of cancer that originates from melanocytes, the pigmentproducing cells derived from the neural crest [4,5]. Despite representing a mere 4% of all skin cancers, CM accounts for up to 75% of skin cancer-related deaths [5]. However, with early detection and proper intervention, over 90% of the cases can be cured [4]. The global estimate of new melanoma cases and deaths is about 325,000, and 57,000, respectively [5-7]. If the rates remain stable, the global burden from melanoma is estimated to increase to 510,000 new cases and 96,000 deaths by 2040 [6]. Despite the increasing number of global melanoma cases, many deaths can be averted through effective prevention, early detection, and effective curative treatments [7]. In Italy, the increasing CM incidence trend is reported to be accurate [8] and increased patient presentation at dermatologic offices has led to increased number of biopsies [8]. Tumor thickness appears to be the most important factor for prediction of malignancy [9]. The survival of patients with melanoma were not consistently associated with mitotic activity. High mitotic activity was generally associated with thick lesions and poor prognosis [9]. One diagnostic problem is that other pigmented lesions including seborrheic keratosis and pigmented basal cell carcinoma can be misdiagnosed as melanomas. Seborrheic Keratosis (SK) is one of the most common epidermal tumors that affects both sexes equally, and usually arises in individuals older than 50 years [10]. It can be similar in gross morphology to melanoma [10]. The tumor may become cancerous in a small fraction of cases [11]. While most dermatologists can identify SK lesions it is sometimes difficult to differentiate between inflammed SK and melanoma [12]. Dermoscopy improves the diagnosis of benign and malignant cutaneous neoplasms in comparison with examination with the unaided eye and should be used routinely for all pigmented and non-pigmented cutaneous neoplasms [12]. 

Optical Coherence Tomography (OCT) has also been used to identify skin cancers. OCT is an imaging technique that applies infrared light to the skin, which allows light to penetrate without causing tissue changes [13-16]. It is used in dermatology for evaluation of benign and cancerous lesions [16]. OCT uses infrared light reflected from the different components of surface tissues to generate an image. It provides cross-sectional and volume scans of skin at depths between 0.4 and 2.0 m with resolution between 3 and 15 micrometers [13-16]. OCT provides quick and useful diagnostic images for several clinical problems and is a valuable addition or complement to other noninvasive imaging tools such as dermoscopy, high-frequency ultrasound, and confocal laser scanning microscopy [17]. While OCT images alone are useful for visual interpretation of skin lesions, additional quantitative information is contained in the images such pixel intensity differences between different skin lesions [18]. Clinical applications of OCT include detection of nonmelanoma skin cancer and evaluation of therapy for inflammatory and connective tissue diseases [19,20]. It also has been reported to be used to evaluate malignant melanomas, basal cell, and squamous cell carcinomas [19,20]. OCT imaging has the potential to serve as an objective, non-invasive measure of disease progression for use in both research trials and clinical practice [21]. The purpose of this paper is to present pilot quantitative pixel intensity vs. depth data and machine learning results obtained from OCT images of melanomas and SKs. Our results show that this information can be used to help noninvasively screen to differentiate between SKs and melanomas.