ABSTRACT
CONTENTS 7.1 Introduction ....................................................................................... 212
7.1.1 Artifact Removal .................................................................... 212 7.1.2 In-Painting.............................................................................. 213 7.1.3 Lesion Segmentation............................................................... 213 7.1.4 Feature Extraction and Classification..................................... 214
7.2 Methods ............................................................................................. 218 7.2.1 Image Hessian Matrix............................................................. 218 7.2.2 Tubularness Filter for Streak Enhancement ........................... 219 7.2.3 Quaternion Tubularness ......................................................... 220 7.2.4 Flux Analysis of the Streaks’ Principal Curvature Vectors .... 220 7.2.5 Streak Detection Features ...................................................... 221
7.3 Machine Learning for Streak Classification........................................ 223 7.4 Results ............................................................................................... 224
T&F Cat #K23910 — K23910 C007 — page 212 — 7/21/2015 — 10:30
7.5 Summary............................................................................................ 225 Acknowledgments ...................................................................................... 226 References .................................................................................................. 226
7.1 INTRODUCTION Malignant melanoma (MM) is one of the common cancers among the white population [1]. Dermoscopy, a noninvasive method for early recognition of MM, allows a clear visualization of skin internal structures, which are often analyzed by a dermoscopic algorithm, such as the ABCD rule of dermoscopy or the 7-point checklist [2]. These methods utilize different dermoscopic features for diagnosing pigmented melanocytic lesions; for example, ABCD analyzes the weighted features of asymmetry (A), border (B), color (C), and differential structures (D). On the other hand, the 7-point checklist looks for the presence of seven different patterns (atypical pigment network, blue-white veil, atypical vascular pattern, irregular streaks, irregular dots/globules, irregular blotches, regression structures). Depending on the presence or absence of each of these patterns, a weight score is assigned to a pigmented lesion. These scores are added up and used for the diagnosis of melanoma. Studies showed that these dermoscopic algorithms can improve the diagnostic accuracy of melanoma.
