Segmentation of skin lesion using Cohen–Daubechies–Feauveau biorthogonal wavelet
© The Author(s) 2016
Received: 11 May 2016
Accepted: 2 September 2016
Published: 19 September 2016
This paper presents a novel technique for segmentation of skin lesion in dermoscopic images based on wavelet transform along with morphological operations. The acquired dermoscopic images may include artifacts inform of gel, dense hairs and water bubble which make accurate segmentation more challenging. We have also embodied an efficient approach for artifacts removal and hair inpainting, to enhance the overall segmentation results. In proposed research, color space is also analyzed and selection of blue channel for lesion segmentation have confirmed better performance than techniques which utilizes gray scale conversion. We tackle the problem by finding the most suitable mother wavelet for skin lesion segmentation. The performance achieved with ‘bior6.8’ Cohen–Daubechies–Feauveau biorthogonal wavelet is found to be superior as compared to other wavelet family. The proposed methodology achieves 93.87 % accuracy on dermoscopic images of PH2 dataset acquired at Dermatology Service of Hospital Pedro Hispano, Matosinhos, Portugal.
Skin cancer is one of a major concern of world due to its alarming increasing rate (Day and Barbour 2000). The most common type of skin cancer is known as melanoma which originate in the pigment-producing melanocytes in bottom layer of skin. Melonoma mostly occurs on sun-exposed parts of the body because intermittent sun exposure damages the skin cells’ DNA with the ultraviolet (UV) radiation, as a result melanocytes become cancerous. However, it can also occur inside eye in iris and also in choroid layer. Melanoma can be classified into malignant melanoma and non-melanoma. Malignant melanoma is less common and accounts only for 5 % of skin cancers however it this is most aggressive, complex and fatal. According to American Cancer Society (ACS), approximately 10,130 people loss their life inside United States in year 2016 because of malignant melanoma (Siegel et al. 2016). The prevalence of melanoma raises every year. Melanoma detection of patients at early stage is of prime importance for the effective treatment. In later stages treatment become hard and melanoma can be fatal.
Segmentation is one of the key step in CAD. In case of skin cancer detection, purpose of segmentation is to segment the effected part of the skin from the normal part. Better segmentation leads better features extraction, classification and diagnosis. Thus diagnostic system’s final results are highly dependent on the quality of segmentation performed. Segmentation could be fully automated (automated analysis end-to-end) or semi-automated (with some mouse clicks). Fully automated segmentation is a complex and challenging task to perform due to the versatility of dermoscopic images from multiple sources, variation of skin color and multiple artifacts like skin tones, hairs, gel, water bubbles or skin lines. The impact of these artifacts is minimized by using image pre-processing steps otherwise already mentioned artifacts can have negative impact on the feature computation and effects skin cancer classification. After removal of undesired artifacts, segmentation process extracts region of interest from the image. Segmentation breakup in whole CAD process is illustrated in Fig. 2.
In this paper, we present our novel segmentation approach for extraction of lesion from skin in presence of artifact like hairs, water bubbles and vessels by employing luminance enhancement and contrast stretching techniques along with wavelet transform for effective segmentation.
Clustering algorithm groups set of pixels in such a ways that pixels in a group are more similar to each other than the pixel in other groups. Lee and Chen (2014) presented an approach based on fuzzy c-mean clustering (FCM) by using type-2 fuzzy set algorithm (Zadeh 1975). They also utilized the 3D color constancy algorithm to minimize the affects of skin tone variations and shadows in images at pre-processing stage. In another approach, fuzzy c-mean clustering and density based clustering (DBSCAN) is utilized for the segmentation of lesion from background skin on mobile platforms (Mendi et al. 2014).
Simple region-based segmentation method groups or split the adjacent pixels or sub-groups into larger or smaller groups on the basis of some homogeneity criteria. The criteria may be color, texture or average gray levels (Gonzalez and Woods 2002). Region base algorithms are not easy to implement as skin lesion have different variable artifacts like different skin type, water bubbles, skin color variation and hairs, which leads to over segmentation (Gonzalez and Woods 2002). Many region based algorithms have been proposed, which includes multi-scale region growing (Hoffmann et al. 2003), morphological flooding (Soille 2013) and statistical region merging (Lissner and Urban 2012). A detailed comparison of different techniques for segmentation of lesions for dermoscopic images is presented in Gómez et al. (2008). This comparison includes techniques based on thresholding and region based methods. However, they excluded edge-based techniques in their comparative analysis.
Edge-based lesion segmentation methods are generally based on the detection of continuous boundary around the lesion using dynamic contour models or with edge detection algorithms. The edge detection algorithms detects image gradients variation to segment lesion from skin (Gonzalez and Woods 2002; Celebi et al. 2008). Abbas et al. (2011) has presented a technique for lesion boarder detection. In their approach, they utilized least-squares method for edge point detection and dynamic programming(DP) to located the boundary of lesion. Another edge based approach using zero-crossings of Laplacian-of-Gaussian (LOG) is presented in Gonzalez and Woods (2002). There are also a large variety of proposed contour based methods (Day and Barbour 2000; Celebi et al. 2008). A contour based technique based on gradient vector flow (GVF) is presented in Day and Barbour (2000). This method is an extension of active contour or normal snake methods, in which curve is deform by the given energy function. Also, Celebi et al. (2008) presented an approach which uses geodesic edge tracing mechanism to locate the active contour of lesion. Most recently, Abuzaghleh et al. (2015) performs lesion segmentation using active contour algorithm along with parse-field level-set method (Whitaker 1998) for active contour evolution. The performance of edge based methods suffers from the presence of fake edges due to artifacts like hairs, vessels and gel in dermosocpic images. Also in some cases boundary between lesion and background skin is not well not define because of smooth transition between skin and lesion (Schmid 1999).
Input image to system
The first step in image segmentation is to prepare the image for segmentation. This is mostly done by applying some pre-processing techniques on the dermoscopic images. The proposed pre-processing stage involves several steps which are described below.
RGB with highest entropy
L*a*b color selection
L*a*b color space by breaking it into L, a and b component has also been experimented during analysis. As L*a*b is representation of CIE 1976 (L*, u*, v*) color space where L represent Lightness and a and b are color-opponent dimensions.1
Blue color selection
This phase analyzes the RBG color space of dermoscopic images for the segmentation of lesion from skin. After the details analysis, color enhancement technique based on blue components is selected for further processing, which gives better segmentation results as compare to other techniques. Therefore blue component is used only because of clear color segmentation between the lesion and normal skin.
Skin hairs frequently appears in dermoscopic images on background skin. Also, hairs partly covered the lesions which causes interference in reliable lesion segmentation. Therefore, hairs should be detected and excluded from dermoscopic image before the inception of skin lesion segmentation procedure. The hair removal process involves three steps i.e. hair enhancement, hair segmentation and hair in-painting. There are number of hair removing methods discussed in literature (Abbasi et al. 2004). During the experiments it has been found that simple morphological operations and directional filter are simple techniques to be used for the hair removal. Moreover, for morphological processing there is a tradeoff between the image edge blurring and the size of structuring elements (Gonzalez and Woods 2002). Due to this morphological trade-off, hair detecting based on direction filter gives better results.
Detection of four dark corner
Segmentation using wavelet transform
Approximation is the image approximation remaining after removing the details. Then details of image is further divided into the horizontal details, vertical details and diagonal details. Approximation can further divided to to next level of approximation and details as shown in Fig. 9. Where H1 gives the horizontal detail, D1 gives the diagonal detail, V1 gives the vertical detail and A1 gives the approximation detail which could be further decomposed to next level. This property of wavelet could be used for image segmentation in medical imaging.
In this work, detail qualitative analysis is performed for the selection of suitable mother wavlet family. The Cohen–Daubechies–Feauveau biorthogonal wavelet is selected and applied on the blue channel of pre-processed image because its demonstrate superiority as compared to other mother wavelet families. During experimentation second level approximate wavelet component gives best results on the inputted image. Although, different combinations are tried by combining two components with different orientation but it has been found that the best results are being obtain through approximation.
After the wavelet transformation, post processing operations are performed to find the final segmented binary result, by keeping large connected binary objects and joining adjacent binary regions. As the processed image at this stage may contain holes due to the intensity difference in skin lesion image. Therefore, morphological operations are performed to fill holes and remove any extra elements other than the skin. The regions belongs to the dark corners around the image is removed by the binary mask in pre-processing step. However the small isolated islands are kept and joined together if they are very near to skin lesion. On the other hand, islands far away from skins lesion are removed by morphological erosion and dilation operations.
Result and discussion
This section contains the detail of different experiments, their setup and results. Various experiments are performed to find out the effective approach for the segmentation of lesion from skin in the presence of undesired artifacts like gel, hairs and water bubbles. Here we present the dermoscopic images dataset and their inclusion criteria including base line metrics. Afterwords experiments performed and relevant procedure adopted during those experiments will be discussed.
The initial requirement of experiment was identification of reliable source for dermatology images. Thus, dermoscopic images for analysis are acquired from Hospital Pedro Hispano database which are 8 bit RGB images of size 768 × 560 pixels. The ground truth is also available with manual segmentation for the comparison with automated segmentation. For the selection between different options in experiment 1, we utilized 45 images out of 200 images. However, segmentation of dermoscopic images by using proposed technique is done in experiment 2 and experiment 3 is performed to compare the acquired result with already existing approaches.
In proposed work, three baseline matrices are used for the qualitative analysis and comparison. These three metrics are: average true detection rate (ATDR), average false positive rate (AFPR), and average error probability rate (AEP).
Experiment 1: Qualitative analysis for the selection of image enhancement approach
Qualitative analysis for gray level selection
RGB color selection
Simple gray conversion
R value from RGB scale
G value from RGB scale
B value from RGB scale
L value from L*a*b scale
a value from L*a*b scale
b value from L*a*b scale
First image is tested with simple gray level conversion with different thresholds, then image RGB color space is analyzed by applying the same threshold value. After that image is converted into the L*a*b color space and each color component is tested separately. The analysis conducted in experiments shows that B values gives the better results as compared to other techniques as well as RGB scales as depicted in Table 1.
Experiment 2: Qualitative analysis for the selection of wavelet family
‘haar’ or ‘db1’ Haar
‘db4’ 4th order Daubechies
‘sym4’ 4th order Symlets
‘bior6.8’ Cohen–Daubechies–Feauveau biorthogonal
Qualitative analysis for wavelet selection
Experiment 3: Qualitative analysis for segmentation of skin lesion
Comparison of proposed method with existing approaches
Discreate wavelet transformation
Adaptive thresholding (AT)
Gradient vector flow (GVF)
Level set method of Chan et al. (C-LS)
Fuzzy-based split-and-merge (FBSM)
Some of the segmentation results of proposed approach on dermoscopic images in presence artifacts like water bubble, hairs and gel effect are shown in Fig. 12. The red outline around the lesion represents the manual segmentation performed by dermatologist while blue outline shows the segmentation achieved by the proposed approach. The segmentation results from the figure clearly demonstrate the effectiveness of proposed approach in the presence of artifacts like veins, hairs, water bubble and gel effects. However, contrast between skin and lesion with similar level of color and intensity variations may cause under and over segmentation which is insignificant. Also artifacts like water bubbles and gel effect in dermoscopic images are removed automatically during segmentation stage with the use of wavelet transformation.
In this paper, we have proposed a DWT based approach for the segmentation of nevus and melanocytic lesions from background skin using dermoscopic images. The analysis is carried to select the appropriate image enhancement technique for segmentation problem. In this regard, techniques like color Entropy, luminance transformation and channels like L*a*b color space with RGB color space are analyzed. However, blue channel is found to appropriate for further processing. Also the proposed technique caters the problem of most unwanted artifacts like hairs and small vessels which are enhanced by using line directional filters and then PDE based in-painting algorithm is employed on detected pixels representing these artifacts. The four dark corners in dermoscopic images are also removed to because it improves segmentation results significantly. In addition to this, segmentation is carried out using Cohen–Daubechies–Feauveau Biorthogonal Wavelets because it produces better results as compared to others like Haar, db4 or sym4. The most significant outcome of using wavelets is the removal of certain artifacts with less effort, e.g. hairs, bubbles and skin tones. The segmented lesion image is enhanced using morphological operations in post processing stage. The proposed methodology is tested on dermoscopic images of PH dataset and achieves ATDR of 93.87 % and AFPR 5.43 %. The experimental results of proposed approach shows the effectiveness of wavelet based segmentation for lesion from background skin.
In order to achieve the accuracy in CAD systems for skin lesion larger dataset is needed. Secondly classification of skin lesions is also not mentioned. Ground truth of images also need to be performed by multiple doctors to reduce its subjectivity and the final ground truth should be established based on these multiple varying judgments.
In case of segmentation, hybrid techniques can be introduced by combining wavelet based segmentation approach with other techniques. Secondly, some new concepts are also emerging such as curvelets which can be explored for segmentation problems. Moreover, better technique of hair detection and removal can enhance the segmentation results.
The research article was designed, directed and coordinated by SK. His conceptual and technical guidance made clear all the aspects of research. UJ worked on the phase of pre-processing of skin lesion images that includes illumination correction, working in different color spaces, noise removal and post processing that leads towards good segmentation of the lesion. UA worked on hair artifact removal that includes highlighting the hairs and then performing inpainting to remove them. KS applied different wavelets and did the segmentation of the images. WM helped in designing the algorithms and implementation. The manuscript was written by WA and KS and commented by all the authors to give it an excellent presentation.
We are really thankful to Higher Education Commission of Pakistan to give the indigenous PhD scholarship to Ms. Uzma Jamil to complete her studies that is the part of this research article. This assignment can not be completed with out the effort and cooperation of all group members.
The authors declare that they have no competing interests.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
- Abbas Q, Celebi ME, Fondón I, Rashid M (2011) Lesion border detection in dermoscopy images using dynamic programming. Skin Res Technol 17(1):91–100View ArticlePubMedGoogle Scholar
- Abbas Q, Fondón I, Rashid M (2011) Unsupervised skin lesions border detection via two-dimensional image analysis. Comput Methods Programs Biomed 3(104):1–15View ArticleGoogle Scholar
- Abbasi NR, Shaw HM, Darrell SR, Darrell SR, Friedman RJ, McCarthy WH, Osman I, Kopf AW, Polsky D (2004) Early diagnosis of cutaneous melanoma: revisiting the ABCD criteria. J Am Med Assoc 292(22):2771–2776View ArticleGoogle Scholar
- Abuzaghleh O, Barkana BD, Faezipour M (2015) Non invasive real-time automated skin lesion analysis system for melanoma early detection and prevention. IEEE J Transl Eng Health Med 3:1–12View ArticleGoogle Scholar
- Castillejos H, Ponomaryov V, Nino-de Rivera L, Golikov V (2012) Wavelet transform fuzzy algorithms for dermoscopic image segmentation. Comput Math Methods Med 2012:578721. doi:10.1155/2012/578721 View ArticlePubMedPubMed CentralMATHGoogle Scholar
- Celebi ME, Hwang S, Iyatomi H, Schaefer G (2010) Robust border detection in dermoscopy images using threshold fusion. In: Proceedings of IEEE international conference on image processing, September 2010Google Scholar
- Celebi ME, Kingravi HA, Iyatomi H, Alp Aslandogan Y, Stoecker WV, Moss RH, Malters JM, Grichnik JM, Marghoob AA, Rabinovitz HS, Menzies SW (2008) Border detection in dermoscopy images using statistical region merging. Skin Res Technol 14(3):347–353View ArticlePubMedPubMed CentralGoogle Scholar
- Day GR, Barbour RH (2000) Automated melanoma diagnosis: where are we at?. Skin Res Technol 6(1):1–15. doi:10.1034/j.1600-0846.2000.006001001.x View ArticlePubMedGoogle Scholar
- Elmisery AM, Rho S, Botvich D (2015) A distributed collaborative platform for personal health profiles in patient-driven health social network. Int J Distrib Sensor Netw 2015:406940. doi:10.1155/2015/406940 Google Scholar
- Gómez DD, Butakoff C, Ersboll BK, Stoecker W (2008) Independent histogram pursuit for segmentation of skin lesions. IEEE Trans Biomed Eng 55(1):157–161View ArticlePubMedPubMed CentralGoogle Scholar
- Gonzalez RC, Woods RE (2002) Digital image processing. Prentice Hall, Englewood CliffsGoogle Scholar
- Hoffmann K, Gambichler T, Rick A, Kreutz M, Anschuetz M, Grünendick T, Orlikov A, Gehlen S, Perotti R, Andreassi L et al (2003) Diagnostic and neural analysis of skin cancer (DANAOS). A multicentre study for collection and computer-aided analysis of data from pigmented skin lesions using digital dermoscopy. Br J Dermatol 149(4):801–809View ArticlePubMedGoogle Scholar
- Huang L-K, Wang M-JJ (1995) Image thresholding by minimizing the measures of fuzziness. Pattern Recogn 28(1):41–51View ArticleGoogle Scholar
- Humayun J, Malik AS, Kamel N (2011) Multilevel thresholding for segmentation of pigmented skin lesions. In: Proceedings of IEEE international conference on imaging systems and techniques, pp 310–314Google Scholar
- Kapur JN, Sahoo PK, Wong AKC (1985) A new method for gray-level picture thresholding using the entropy of the histogram. Comput Vis Graph Image Process 29:273–285View ArticleGoogle Scholar
- Kittler J, Illingworth J (1986) Minimum error thresholding. Pattern Recogn 19(1):41–47View ArticleGoogle Scholar
- Kruk M, Świderski B, Osowski S, Kurek J, Słowińska M, Walecka I (2015) Melanoma recognition using extended set of descriptors and classifiers. J Image Video Process. doi:10.1186/s13640-015-0099-9 Google Scholar
- Lee H, Chen YP (2014) Skin cancer extraction with optimum Fuzzy thresholding technique. Appl Intell 40(3):415–426View ArticleGoogle Scholar
- Lissner I, Urban P (2012) Toward a unified color space for perception-based image processing. IEEE Trans Image Process 21(3):1153–1168ADSMathSciNetView ArticlePubMedGoogle Scholar
- Mendi E, Yogurtcular C, Sezgin Y, Bayrak C (2014) Automatic mobile segmentation of dermoscopy images using density based and fuzzy c-means clustering. In: 2014 IEEE international symposium on medical measurements and applications (MeMeA), pp 1–6, June 2014Google Scholar
- Otsu N (1975) A threshold selection method from gray-level histograms. Automatica 11(285–296):23–27Google Scholar
- Qian X, Wang J, Guo S, Li Q (2013) An active contour model for medical image segmentation with application to brain CT image. Med Phys 40(2):021911View ArticlePubMedPubMed CentralGoogle Scholar
- Sadri AR, Zekri M, Sadri S, Gheissari N, Mokhtari M, Kolahdouzan F (2013) Segmentation of dermoscopy images using wavelet networks. IEEE Trans Biomed Eng 60(4):1134–1141View ArticlePubMedGoogle Scholar
- Schmid P (1999) Segmentation of digitized dermatoscopic images by two-dimensional color clustering. IEEE Trans Med Imag 18(2):164–171View ArticleGoogle Scholar
- Siegel RL, Miller KD, Jemal A (2016) Cancer statistics, 2016. CA Cancer J Clin 66(1):7–30View ArticlePubMedGoogle Scholar
- Silveira M, Nascimento JC, Marques JS, Maral AR, Mendonca T, Yamauchi S, Maeda J, Rozeira J (2009) Comparison of segmentation methods for melanoma diagnosis in dermoscopy images. IEEE J Sel Top Signal Process 3(1):35–45ADSView ArticleGoogle Scholar
- Soille P (2013) Morphological image analysis: principles and applications. Springer, BerlinMATHGoogle Scholar
- Vestergaard ME, Macaskill PHPM, Holt PE, Menzies SW (2008) Dermoscopy compared with naked eye examination for the diagnosis of primary melanoma: a meta-analysis of studies performed in a clinical setting. Br J Dermatol 159(3):669–676PubMedGoogle Scholar
- Whitaker RT (1998) A level-set approach to 3d reconstruction from range data. Int J Comput Vis 29(3):203–231View ArticleGoogle Scholar
- Zadeh LA (1975) The concept of a linguistic variable and its application to approximate reasoning I. Inf Sci 8(3):199–249MathSciNetView ArticleMATHGoogle Scholar