Thesis Proposal: Merve Aydin Chester (AYDMER001)
Pre-processing and Segmentation of Skin Lesion Images
Using Watershed and k-means Algorithms
Malignant melanoma, the most deadly form of skin cancer, is one of the
most rapidly increasing cancers in the world, with an estimated incidence
of 68,720 and an estimated total of 8650 deaths in the United States in
2009 aloneis generally used for discovering clusters in large
spatial databases wit h noise
Ester M, Kriegel HP, Sander J, Xu X: A density-based algorithm for
discovering clusters in large spatial databases with noise. Conference on
Knowledge Discovery and Data Mining 1577350049 1996, DOI: [Link].1980.
Introduction
Malignant melanoma is the most deadly form of skin cancer, and it is also one of the
most rapidly increasing cancers in the world. In the United States in 2009 alone 68,720
were diagnosed with melanoma and 8,650 people died because of it (Jamal et al., 2009).
In Europe, skin cancer accounted for 35,000 new cases and 9,000 deaths in 2000 (Boyle
et al., 2004). Early diagnosis of melanoma can help save lives as it is easy to cure with a
surgical excision when detected early.
Background
One of the most important tools to diagnose melanoma is dermoscopy, which helps
clinicians monitor skin lesions that are potentially cancerous. If the images captured
from two different patient visits show that the lesion remains the same over time, it is
most probably benign. If the lesion changes, then it may require excision.
Dermoscopy is a non-invasive skin imaging technique that involves optical
magnification, along with optics that minimize surface reflection, making subsurface
structures more easily visible compared to conventional clinical images (Argenziano,
Soyer and De Giorgi, 2002). Visual interpretation using dermoscopy is highly effective in
identifying the features of melanomas (Menzies et al., 2009). However, its accuracy is
much lower when practiced by less experienced clinicians (Binder et al, 1995).
Computerized image analysis is of great importance, therefore, in reducing diagnostic
errors resulting from the subjectivity and difficulty of visual interpretation (Fleming et
al., 1998).
Segmentation is the term used for separating an image into meaningful areas and it is
the first and crucial step for analysing dermoscopy images. There is a lot of research
that has been done on this topic. Celebi at al (2005) has a good literature review for pre-
processing methods, segmentation and border detection for skin lesion images.
According to Celebi et al (2005) the methods are separated into nine categories:
1) Histogram thresholding
2) Clustering
3) Edge based
4) Region based
5) Morphological
6) Model based
7) Active contours
Thesis Proposal: Merve Aydin Chester (AYDMER001)
Pre-processing and Segmentation of Skin Lesion Images
Using Watershed and k-means Algorithms
8) Fuzzy Logic
9) Supervised learning
Thesis Proposal: Merve Aydin Chester (AYDMER001)
Pre-processing and Segmentation of Skin Lesion Images
Using Watershed and k-means Algorithms
Thesis proposal
In previous research, analysts have used k-means as a method of segmentation and
classification. In this paper I will also use k-means for segmentation, but will apply a
number of pre-processing steps and will implement a watershed algorithm before then
connecting various segments using k-means clustering.
In this thesis I will test the following sequence of pre-processing steps and
segmentation of skin lesion images using the watershed algorithm and k-means
clustering
Pre-processing steps: Image scaling, then elimination correction, and finally hair
removal.
Segmentation steps: We will implement a watershed algorithm followed by k-
means clustering. The watershed algorithm suffers from over-segmentation, so
the segmented regions which are found after application of the watershed
algorithm will be clustered using the k-means algorithm.
After all the aforementioned steps are complete, the accuracy of the segmentation will
be calculated using the Jacard index, and sensitivity and precision measures.
References and further reading
1. Jemal, A., Siegel, R., Ward, E., Hao, Y., Xu, J. and Thun, M. (2009). Cancer Statistics,
2009. CA: A Cancer Journal for Clinicians, [online] 59(4), pp.225-249. Available at:
[Link] [Accessed 26 May 2019].
2. Boyle, P., Doré, J., Autier, P. and Ringborg, U. (2004). Cancer of the skin: a
forgotten problem in Europe. Annals of Oncology, 15(1), pp.5-6.
3. Argenziano, G., Soyer, H. and De Giorgi, V. (2002). Dermoscopy: a tutorial. Milan:
EDRA Medical Publishing & New Media.
4. Menzies, S., Crotty, K., Ingwar, C. and McCarty, W. (2009). Dermoscopy: An atlas.
New York: McGraw-Hill Companies.
5. Binder, M., Schwarz, M., Winkler, A., Steiner, A., Wolff, K. and Pehamberger, H.
(1993). Application of epiluminescence microscopy in the diagnosis of
pigmented skin lesions is a questionable tool for not specifically trained
dermatologists. Journal of Dermatological Science, [online] 6(1), p.101. Available
at: [Link]
6. Fleming, M., Steger, C., Zhang, J., Gao, J., Cognetta, A., Pollak, l. and R. Dyer, C.
(1998). Techniques for a structural analysis of dermatoscopic
imagery. Computerized Medical Imaging and Graphics, [online] 22(5), pp.375-
389. Available at: [Link]
7. Celebi, M., Wen, Q., Iyatomi, H., Shimizu, K., Zhou, H. and Schaefer, G. (2015). A
State-of-the-Art Survey on Lesion Border Detection in Dermoscopy
Images. Dermoscopy Image Analysis, [online] pp.97-129. Available at:
[Link]
Art_Survey_on_Lesion_Border_Detection_in_Dermoscopy_Images [Accessed 26
May 2019].
