ABSTRACT
CONTENTS 11.1 Introduction ..................................................................................... 346
11.1.1 Preprocessing: Removal of Artifacts .................................... 347 11.1.2 Skin Lesion Segmentation .................................................... 348 11.1.3 Feature Extraction............................................................... 349 11.1.4 Feature Selection.................................................................. 350 11.1.5 Classification........................................................................ 350
11.2 Proposed System .............................................................................. 351 11.2.1 Hair Removal Algorithm...................................................... 351
11.2.1.1 Hair Detection from Dermoscopy Images.............. 351 11.2.1.2 Hair Repair............................................................ 351
11.2.2 Lesion Border Detection in Dermoscopy Images.................. 353 11.2.3 Feature Extraction............................................................... 354
11.2.3.1 Geometric Features ............................................... 355 11.2.3.2 Color Features ....................................................... 357 11.2.3.3 Texture Features ................................................... 357 11.2.3.4 Wavelet-Based Texture Features ........................... 360
11.2.4 Features Normalization ........................................................ 361
T&F Cat #K23910 — K23910 C011 — page 346 — 7/14/2015 — 9:35
11.2.5 Feature Selection.................................................................. 361 11.2.5.1 Correlation-Based Feature Selection ..................... 361 11.2.5.2 Relief-Based Feature Selection .............................. 362 11.2.5.3 T-Test Feature Selection ....................................... 362
11.2.6 Classification........................................................................ 362 11.2.6.1 k-Nearest Neighbor................................................ 362 11.2.6.2 Support Vector Machine........................................ 363 11.2.6.3 Artificial Neural Network ...................................... 363
11.3 Experimental Results and Discussion............................................... 364 11.3.1 Hair Removal ....................................................................... 364
11.3.1.1 Assessment of the Proposed Hair Detection Method .................................................................. 364
11.3.1.2 Assessment of theProposedHairRepairMethod..... 365 11.3.2 Lesion Segmentation ............................................................ 365
11.3.2.1 Comparison with Other Automated Segmentation Methods .......................................... 369
11.3.2.2 Effect of Proposed Hair Removal Method on Lesion Segmentation.............................................. 370
11.3.3 Feature Extraction, Feature Selection, and Classification.... 370 11.3.3.1 Selection of Optimal Features ............................... 372 11.3.3.2 Evaluation of KNN Classification Algorithm ........ 373 11.3.3.3 Evaluation of SVM Classification Algorithm......... 373 11.3.3.4 Evaluation of ANN Classification Algorithm ........ 374 11.3.3.5 Comparison of Different Classification Methods ... 375
11.4 Conclusion ........................................................................................ 376 References .................................................................................................. 377
11.1 INTRODUCTION Melanoma is a very serious form of cancer that occurs most often in the skin; it is also known as cutaneous melanoma. The word melanoma is derived from the Greek words melas (black) and -oma (tumor). Melanoma is initiated in melanocyte cells that produce a pigment called melanin. The major environmental risk factor for melanoma is overexposure to the sun’s harmful rays, known as ultraviolet (UV) radiation. Melanoma is the least common of all skin cancers, but is the most deadly type. According to the American Cancer Society, melanoma accounts for only about 4% of all skin cancer cases but is responsible for 79% of all skin cancer-related deaths [1]. Melanoma is now the seventh most frequent cancer in Canada and the fifth most common malignancy in the United States [2]. The good news is that when melanoma is diagnosed in its early stages, it can be treated and cured without complications and the chances for long-term, disease-free survival are excellent. Nevertheless, the early diagnosing of melanoma is not always trivial even for
T&F Cat #K23910 — K23910 C011 — page 347 — 7/14/2015 — 9:35
experienced dermatologists and more probably for primary care physicians or less experienced dermatologists [3, 4].