Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Hair removal from dermoscopic color images

Hair removal from dermoscopic color images Skin cancer is the most commonly diagnosed type of cancer in people, regardless of age, gender, or race. One of the most common malignant skin cancers is melanoma, which is a dangerous proliferation of melanocytes. It is a well-known fact that early diagnosis of skin cancer is crucial and allows for successful treatment. Treatment of melanoma is not effective when melanoma is at an advanced stage. A widely used tool for the examination of skin lesions is a dermatoscope, which uses optic magnification to visualize features that are invisible to the naked eye. For a precise and objective diagnosis, there is a need for a computerized method for the removal and inpainting of hairs in image processing. In this study, we present an algorithm for the detection and inpainting of hairs in color dermoscopic images. Keywords: dermoscopy; hair removal; image inpainting; skin cancer; top-hat transform. and diagnoses and death rates are increasing faster than any other skin cancer. One of the major contributors to the development of melanoma is ultraviolet radiation (long-term sun exposure and sun burn), which causes damage to cell DNA. In addition, the negative influence of quality of life is of great importance. Owing to increased skin cancer incidence, dermatological oncology has become a fast developing branch of medicine. One of the main tasks of modern dermatology is the detection of melanoma in its early stage of development, because the survival rate after identification of < 0.75 mm thick melanomas is near 100% [1, 2]. Dermoscopy (also known as dermatoscopy or epiluminescence microscopy) is a noninvasive, in vivo medical examination that uses optic magnification to visualize features of the pigmented skin lesion that are invisible to the naked eye (Figure 1) [1, 2]. Physicians assess the dermoscopic image based on the presence or absence of different global and local features. The most widely used diagnostic algorithms are the ABCD rule described in 1993 by Stolz et al. and the seven-point checklist proposed by Argenziano and colleagues in 1998 [1]. The early detection of melanoma can be achieved by a well-developed computer diagnostic system based on image processing and classification methods. One of the most important steps is the preprocessing stage, that is, an improvement of the image data that suppresses undesired distortions or enhances some image features relevant for further image analysis [5]. Skin lesion images often contain extraneous noise such as skin texture, air bubbles, and hair that make border and feature detection more difficult. To reduce the effects of noise, images should be preprocessed. One of the most important parts of this step is the removal and restoration of hair and hair-like regions within skin lesion images (Figure 2). In dermoscopic images, two types of hairs can occur: small and light as well as thick and dark. Hairs are generally long, straight curvilinear structures with relatively constant width and curvature. This paper is organized as follows. First, the preprocessing algorithm, including black frame removal, filtering, hair removal, and inpainting, are described. Then, the conducted tests and results are described. Finally, we close the paper with concluding remarks, discuss the results, and highlight future directions. *Corresponding author: Joanna Jaworek-Korjakowska, Department of Automatics and Biomedical Engineering, AGH University of Science and Technology, Biocybernetics Laboratory, al. A. Mickiewicza 30, 30-058 Krakow, Poland, E-mail: jaworek@agh.edu.pl Ryszard Tadeusiewicz: Department of Automatics and Biomedical Engineering, Biocybernetics Laboratory, AGH University of Science and Technology, al. A. Mickiewicza 30, 30-058 Krakow, Poland Introduction Skin cancer is the most commonly diagnosed type of cancer in people, regardless of age, gender, or race. One of the most common malignant skin cancers is melanoma, which is a dangerous proliferation of melanocytes. Malignant melanoma (Latin: melanoma malignum) originates in pigment producing cells called melanocytes, which derive from the neural crest. This type of tumor occurs mainly in the skin, but it can also be found in mucous membranes of the gastrointestinal tract and even in the eye [1­4]. Melanomas are fast-growing and highly malignant tumors often spreading to nearby lymph nodes, lungs, and the brain. In the past several years, an increasing incidence of melanoma has been observed worldwide 54Jaworek-Korjakowska and Tadeusiewicz: Hair removal from dermoscopic color images Hair detection and removal algorithm The main aim of the algorithm is to repair the texture of the skin lesion, which becomes consistent, and prepare the image for the segmentation step, feature extraction, and classification. For this reason, an automatic preprocessing system has been developed and is based on the flowchart of the algorithm demonstrated in Figure 3. The approach is divided into four stages: black frame removal (Figure 4), filtering, hair removal, and inpainting. The small hairs are removed in the filtering process, whereas the thick hairs are removed with the described hair removal algorithm. Black frame removal In the first step, we remove the black frame which is introduced during the digitization process (normal output from some types of dermoscopy). To determine the darkness of a pixel with (R,G,B) coordinates, the lightness component of the HSL color space [6] is calculated as Eq. (1): Figure 1Lesion observed with the naked eye in comparison to dermoscopic examination. Global and local features become visible (based on [1]). Figure 2Dermoscopic images with hair pixels. Different examples: (A), (B), (C), and (D) [1]. Jaworek-Korjakowska and Tadeusiewicz: Hair removal from dermoscopic color images55 Medical image Black frame removal Gaussian filter Grayscale image Hair removal High-pass filter Inpainting Top-hat transform Eliminate unwanted pixels Figure 3Preprocessing and hair removal algorithm. L= max( R, G , B ) + min( R, G , B ) 2 (1) 1 G( x , y ) = e 2 2 x 2 +y 2 2 2 (2) A pixel is considered to be black when the lightness value is < 15 (range of the lightness value [0:255]). We scan the rows in four directions (top, bottom, right, left) and calculate the amount of black pixels. A particular row is labeled as part of the black frame if it contains 50% black pixels and is removed. Additionally, we remove ten more rows that represent the light-colored part of the frame (Figure 4). Hair removal Small and light hairs are reduced in the filtering step (Figure 5). Removal and restoration of dark, thick hairs and hair-like regions within skin lesion images has to be done separately and is needed for effective segmentation and classification of global and local features. Several methods have been developed for hair removal on dermoscopic images, mostly based on morphological operations and adaptive thresholding [3, 6, 10, 11]. These methods are fast but mostly remove subtle and important features that are misinterpreted as hairs. A good approach for hair removal is the use of top-hat transform. The process consists of four steps. Firstly, we convert the dermoscopic RGB image into grayscale with the standard NTSC conversion equation, Eq. (3) [2]: G(i, j) = 0.2989I(i, j, 1)+0.587I(i, j, 2)+0.114I(i, j, 3) (3) Filtering The purpose of the second step is to reduce noise such as skin lines, air bubbles and light, small hairs around the lesion. To smooth these artifacts we use Gaussian smoothing, which is a low-pass filter. The Gaussian blur is a type of image-blurring filter that uses a Gaussian function for calculating the transformation to apply to each pixel in the image [7­9]. The equation of a Gaussian function in two dimensions is defined by Eq. (2): Figure 4Black frame removal: (A) dermoscopic image with frame, (B) outcome of the algorithm. 56Jaworek-Korjakowska and Tadeusiewicz: Hair removal from dermoscopic color images Figure 5Filtering stage: (A) part of the dermoscopic image, (B) result of the filtering process. In the second step, a generalization of high-pass filter (unsharp masking), based on the negative of the Laplacian filter, is used. Unsharp masking yields increased local contrast. In the third stage, a black top-hat transform, which detects hairs in the image, is performed. Top-hat transform is an operation that extracts small elements and details from given images [7, 10]. The black top-hat transform of f is given by Eq. (4): BTH(f) = B(f)­f (4) analysis of hair detection quality was based on the evaluation of diagnostic accuracy (DA), Eq. (5), as well as sensitivity (SE), Eq. (6). DA = TP TP + FN + FP TP TP + FN (5) SE = (6) where B(f) is a closing operation. This method removes the background while preserving the hair strands, regardless of the nature of the neighbor areas. In the last step, we refine the broken lines and remove unwanted pixels by calculating the area, circularity, major and minor axis. The aim of the last step is to inpaint the detected hair lines. Hair line pixels are replaced with values calculated on the basis of the neighborhood pixels (Figure 6). where TP (true positive) denotes pixels correctly marked as hairs; FN (false negative) denotes pixels incorrectly marked as background; and FP (false positive) denotes pixels incorrectly marked as hairs. The hair removal algorithm gave 88.7% DA and 90.8% SE. Discussion and conclusions The results indicate that the proposed algorithm can be used as a preprocessing step and hair removal in dermoscopic color images. The proposed algorithm could be part of a whole melanoma diagnostic system that would be used not only by young inexperienced dermatologists but first and foremost by family physicians [12, 13]. This is an opportunity for people that live in remote and rural areas outside regional centers and are faced with the usual difficulties of making an appointment with a dermatologist. It is very important to diagnose melanomas at an early stage because early diagnosis will reduce the melanomarelated mortality rate [12­14]. Despite the fact that the results are satisfactory, the proposed algorithm will still need to be developed and compared with other algorithms. Experimental results and analysis The proposed and implemented algorithm for the removal of hairs in dermoscopic images has been tested on over 50 images from two university hospitals (University of Naples, Italy and University of Graz, Austria) that were stored on a CD-ROM in JPEG format [1]. Documentation of the dermoscopic images was performed using a Dermaphot apparatus (Heine, Optotechnik, Herrsching, Germany) or a photo camera (Nikon F3, Japan) mounted on a stereomicroscope (Wild M650, Heerbrugg AG, Heerbrugg, Switzerland) [1]. Every image has been assessed manually and by the described hair removal algorithm. The performance Jaworek-Korjakowska and Tadeusiewicz: Hair removal from dermoscopic color images57 Figure 6Hair removal results after: (A) dermoscopic image acquisition, (B) from RGB to grayscale conversion, (C) unsharp masking, (D) black top-hat transform, (E) after removing unwanted pixels (presented in white), and (F) inpainting. In conclusion, the implemented algorithm meets expectations. The results of the preliminary tests show that image preprocessing can remove noise and artifacts in dermoscopic images and prepare the image for better and precise evaluation [4, 15, 16]. Acknowledgments: This scientific research was supported by the National Science Center as research project no. 2011/01/N/ST7/06783. Received March 18, 2013; accepted April 30, 2013 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Bio-Algorithms and Med-Systems de Gruyter

Hair removal from dermoscopic color images

Loading next page...
 
/lp/de-gruyter/hair-removal-from-dermoscopic-color-images-0cZXo5DQUJ

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
de Gruyter
Copyright
Copyright © 2013 by the
ISSN
1895-9091
eISSN
1896-530X
DOI
10.1515/bams-2013-0013
Publisher site
See Article on Publisher Site

Abstract

Skin cancer is the most commonly diagnosed type of cancer in people, regardless of age, gender, or race. One of the most common malignant skin cancers is melanoma, which is a dangerous proliferation of melanocytes. It is a well-known fact that early diagnosis of skin cancer is crucial and allows for successful treatment. Treatment of melanoma is not effective when melanoma is at an advanced stage. A widely used tool for the examination of skin lesions is a dermatoscope, which uses optic magnification to visualize features that are invisible to the naked eye. For a precise and objective diagnosis, there is a need for a computerized method for the removal and inpainting of hairs in image processing. In this study, we present an algorithm for the detection and inpainting of hairs in color dermoscopic images. Keywords: dermoscopy; hair removal; image inpainting; skin cancer; top-hat transform. and diagnoses and death rates are increasing faster than any other skin cancer. One of the major contributors to the development of melanoma is ultraviolet radiation (long-term sun exposure and sun burn), which causes damage to cell DNA. In addition, the negative influence of quality of life is of great importance. Owing to increased skin cancer incidence, dermatological oncology has become a fast developing branch of medicine. One of the main tasks of modern dermatology is the detection of melanoma in its early stage of development, because the survival rate after identification of < 0.75 mm thick melanomas is near 100% [1, 2]. Dermoscopy (also known as dermatoscopy or epiluminescence microscopy) is a noninvasive, in vivo medical examination that uses optic magnification to visualize features of the pigmented skin lesion that are invisible to the naked eye (Figure 1) [1, 2]. Physicians assess the dermoscopic image based on the presence or absence of different global and local features. The most widely used diagnostic algorithms are the ABCD rule described in 1993 by Stolz et al. and the seven-point checklist proposed by Argenziano and colleagues in 1998 [1]. The early detection of melanoma can be achieved by a well-developed computer diagnostic system based on image processing and classification methods. One of the most important steps is the preprocessing stage, that is, an improvement of the image data that suppresses undesired distortions or enhances some image features relevant for further image analysis [5]. Skin lesion images often contain extraneous noise such as skin texture, air bubbles, and hair that make border and feature detection more difficult. To reduce the effects of noise, images should be preprocessed. One of the most important parts of this step is the removal and restoration of hair and hair-like regions within skin lesion images (Figure 2). In dermoscopic images, two types of hairs can occur: small and light as well as thick and dark. Hairs are generally long, straight curvilinear structures with relatively constant width and curvature. This paper is organized as follows. First, the preprocessing algorithm, including black frame removal, filtering, hair removal, and inpainting, are described. Then, the conducted tests and results are described. Finally, we close the paper with concluding remarks, discuss the results, and highlight future directions. *Corresponding author: Joanna Jaworek-Korjakowska, Department of Automatics and Biomedical Engineering, AGH University of Science and Technology, Biocybernetics Laboratory, al. A. Mickiewicza 30, 30-058 Krakow, Poland, E-mail: jaworek@agh.edu.pl Ryszard Tadeusiewicz: Department of Automatics and Biomedical Engineering, Biocybernetics Laboratory, AGH University of Science and Technology, al. A. Mickiewicza 30, 30-058 Krakow, Poland Introduction Skin cancer is the most commonly diagnosed type of cancer in people, regardless of age, gender, or race. One of the most common malignant skin cancers is melanoma, which is a dangerous proliferation of melanocytes. Malignant melanoma (Latin: melanoma malignum) originates in pigment producing cells called melanocytes, which derive from the neural crest. This type of tumor occurs mainly in the skin, but it can also be found in mucous membranes of the gastrointestinal tract and even in the eye [1­4]. Melanomas are fast-growing and highly malignant tumors often spreading to nearby lymph nodes, lungs, and the brain. In the past several years, an increasing incidence of melanoma has been observed worldwide 54Jaworek-Korjakowska and Tadeusiewicz: Hair removal from dermoscopic color images Hair detection and removal algorithm The main aim of the algorithm is to repair the texture of the skin lesion, which becomes consistent, and prepare the image for the segmentation step, feature extraction, and classification. For this reason, an automatic preprocessing system has been developed and is based on the flowchart of the algorithm demonstrated in Figure 3. The approach is divided into four stages: black frame removal (Figure 4), filtering, hair removal, and inpainting. The small hairs are removed in the filtering process, whereas the thick hairs are removed with the described hair removal algorithm. Black frame removal In the first step, we remove the black frame which is introduced during the digitization process (normal output from some types of dermoscopy). To determine the darkness of a pixel with (R,G,B) coordinates, the lightness component of the HSL color space [6] is calculated as Eq. (1): Figure 1Lesion observed with the naked eye in comparison to dermoscopic examination. Global and local features become visible (based on [1]). Figure 2Dermoscopic images with hair pixels. Different examples: (A), (B), (C), and (D) [1]. Jaworek-Korjakowska and Tadeusiewicz: Hair removal from dermoscopic color images55 Medical image Black frame removal Gaussian filter Grayscale image Hair removal High-pass filter Inpainting Top-hat transform Eliminate unwanted pixels Figure 3Preprocessing and hair removal algorithm. L= max( R, G , B ) + min( R, G , B ) 2 (1) 1 G( x , y ) = e 2 2 x 2 +y 2 2 2 (2) A pixel is considered to be black when the lightness value is < 15 (range of the lightness value [0:255]). We scan the rows in four directions (top, bottom, right, left) and calculate the amount of black pixels. A particular row is labeled as part of the black frame if it contains 50% black pixels and is removed. Additionally, we remove ten more rows that represent the light-colored part of the frame (Figure 4). Hair removal Small and light hairs are reduced in the filtering step (Figure 5). Removal and restoration of dark, thick hairs and hair-like regions within skin lesion images has to be done separately and is needed for effective segmentation and classification of global and local features. Several methods have been developed for hair removal on dermoscopic images, mostly based on morphological operations and adaptive thresholding [3, 6, 10, 11]. These methods are fast but mostly remove subtle and important features that are misinterpreted as hairs. A good approach for hair removal is the use of top-hat transform. The process consists of four steps. Firstly, we convert the dermoscopic RGB image into grayscale with the standard NTSC conversion equation, Eq. (3) [2]: G(i, j) = 0.2989I(i, j, 1)+0.587I(i, j, 2)+0.114I(i, j, 3) (3) Filtering The purpose of the second step is to reduce noise such as skin lines, air bubbles and light, small hairs around the lesion. To smooth these artifacts we use Gaussian smoothing, which is a low-pass filter. The Gaussian blur is a type of image-blurring filter that uses a Gaussian function for calculating the transformation to apply to each pixel in the image [7­9]. The equation of a Gaussian function in two dimensions is defined by Eq. (2): Figure 4Black frame removal: (A) dermoscopic image with frame, (B) outcome of the algorithm. 56Jaworek-Korjakowska and Tadeusiewicz: Hair removal from dermoscopic color images Figure 5Filtering stage: (A) part of the dermoscopic image, (B) result of the filtering process. In the second step, a generalization of high-pass filter (unsharp masking), based on the negative of the Laplacian filter, is used. Unsharp masking yields increased local contrast. In the third stage, a black top-hat transform, which detects hairs in the image, is performed. Top-hat transform is an operation that extracts small elements and details from given images [7, 10]. The black top-hat transform of f is given by Eq. (4): BTH(f) = B(f)­f (4) analysis of hair detection quality was based on the evaluation of diagnostic accuracy (DA), Eq. (5), as well as sensitivity (SE), Eq. (6). DA = TP TP + FN + FP TP TP + FN (5) SE = (6) where B(f) is a closing operation. This method removes the background while preserving the hair strands, regardless of the nature of the neighbor areas. In the last step, we refine the broken lines and remove unwanted pixels by calculating the area, circularity, major and minor axis. The aim of the last step is to inpaint the detected hair lines. Hair line pixels are replaced with values calculated on the basis of the neighborhood pixels (Figure 6). where TP (true positive) denotes pixels correctly marked as hairs; FN (false negative) denotes pixels incorrectly marked as background; and FP (false positive) denotes pixels incorrectly marked as hairs. The hair removal algorithm gave 88.7% DA and 90.8% SE. Discussion and conclusions The results indicate that the proposed algorithm can be used as a preprocessing step and hair removal in dermoscopic color images. The proposed algorithm could be part of a whole melanoma diagnostic system that would be used not only by young inexperienced dermatologists but first and foremost by family physicians [12, 13]. This is an opportunity for people that live in remote and rural areas outside regional centers and are faced with the usual difficulties of making an appointment with a dermatologist. It is very important to diagnose melanomas at an early stage because early diagnosis will reduce the melanomarelated mortality rate [12­14]. Despite the fact that the results are satisfactory, the proposed algorithm will still need to be developed and compared with other algorithms. Experimental results and analysis The proposed and implemented algorithm for the removal of hairs in dermoscopic images has been tested on over 50 images from two university hospitals (University of Naples, Italy and University of Graz, Austria) that were stored on a CD-ROM in JPEG format [1]. Documentation of the dermoscopic images was performed using a Dermaphot apparatus (Heine, Optotechnik, Herrsching, Germany) or a photo camera (Nikon F3, Japan) mounted on a stereomicroscope (Wild M650, Heerbrugg AG, Heerbrugg, Switzerland) [1]. Every image has been assessed manually and by the described hair removal algorithm. The performance Jaworek-Korjakowska and Tadeusiewicz: Hair removal from dermoscopic color images57 Figure 6Hair removal results after: (A) dermoscopic image acquisition, (B) from RGB to grayscale conversion, (C) unsharp masking, (D) black top-hat transform, (E) after removing unwanted pixels (presented in white), and (F) inpainting. In conclusion, the implemented algorithm meets expectations. The results of the preliminary tests show that image preprocessing can remove noise and artifacts in dermoscopic images and prepare the image for better and precise evaluation [4, 15, 16]. Acknowledgments: This scientific research was supported by the National Science Center as research project no. 2011/01/N/ST7/06783. Received March 18, 2013; accepted April 30, 2013

Journal

Bio-Algorithms and Med-Systemsde Gruyter

Published: Jun 1, 2013

References