ABSTRACT
CONTENTS 10.1 Introduction ..................................................................................... 294 10.2 Background ...................................................................................... 297
10.2.1 Image Acquisition ................................................................ 297 10.2.2 Segmentation ....................................................................... 299 10.2.3 Feature Extraction............................................................... 303
10.2.3.1 Asymmetry............................................................ 303 10.2.3.2 Border ................................................................... 306 10.2.3.3 Color...................................................................... 308 10.2.3.4 Differential Structures ........................................... 310 10.2.3.5 High-Level Intuitive Features ................................ 311
T&F Cat #K23910 — K23910 C010 — page 294 — 7/21/2015 — 10:34
10.2.4 Feature Selection.................................................................. 311 10.2.5 Classification........................................................................ 313
10.2.5.1 Discriminant Analysis ........................................... 313 10.2.5.2 Artificial Neural Network ...................................... 314 10.2.5.3 k-Nearest-Neighborhood........................................ 314 10.2.5.4 Support Vector Machine........................................ 314 10.2.5.5 Decision Trees ....................................................... 315 10.2.5.6 Ensemble Classifiers .............................................. 315 10.2.5.7 Comparison of Different Machine Learning
Methods ................................................................ 316 10.3 Methodology: Visual Recognition Approach .................................... 317
10.3.1 Low-Level Visual Features ................................................... 318 10.3.2 Ensemble Modeling Approach.............................................. 320 10.3.3 Large-Scale Training ............................................................ 322
10.4 Experimental Results ....................................................................... 322 10.4.1 Datasets ............................................................................... 323 10.4.2 Pattern Classification........................................................... 323 10.4.3 Disease Classification ........................................................... 324
10.4.3.1 Using Low-Level Feature Only on PH2................. 324 10.4.3.2 Using Pattern Classifier on PH2 ........................... 325 10.4.3.3 Using Ensemble Modeling on ISIC........................ 326
10.5 Conclusion ........................................................................................ 328 References .................................................................................................. 329
10.1 INTRODUCTION Malignant melanoma is the deadliest form of skin cancer; it makes up only 2.3% of all skin cancers, yet is responsible for 75% of skin cancer deaths. The American Cancer Society reported 76,100 new cases of melanoma in the United States in 2013, with 9710 melanoma deaths, maintaining an increasing trend over the last decades [1]. In Australia, melanoma is the most common cancer in people aged 15-44 years. It represents 10% of all cancers and its per-capita incidence is four times higher than in Canada, the UK, and the United States, with more than 10,000 cases diagnosed and around 1250 deaths annually [2]. The worldwide steady increase in incidence of melanoma in recent years [3], its high mortality rate, and the massive respective medical cost have made its early diagnosis a continuing priority of public health.
