Skin Scan Melanoma Screening Using Smart Handheld Devices

Skin. Scan©: Melanoma Screening Using Smart Handheld Devices Tarun Wadhawan 1, Ning Situ 1, Hu Rui 2, Yuriy Fofanov 1, and George Zouridakis 1, 2, 3 Dept. of Computer Science 1, Electrical and Computer Engineering 2, and Engineering Technology 3, University of Houston, Houston Texas, USA INTRODUCTION METHOD • 68, 100 new cases and 8, 700 deaths in U. S. (2010) • Deployment on Apple i. Phone as a screening device for skin cancer detection • Embedded processors have limited processing speed and memory • C/C++ based library for image processing and pattern recognition II. Feature Extraction 1. Texture features § Haar wavelet features based on bagof-features § Local binary pattern: Invariant to change in intensity and rotation 2. Color features § Color histograms: Intensity distribution of HSV, RGB, and LAB color spaces III. Classification Feature selection and linear SVM Irregular Pigment Blue. Whitish Veil Irregular Streaks Irregular Dots Atypical Pigment Network Atypical Vascular Pattern Regression Structures METHOD I. Detection of Region of Interest (ROI): 1. Pre-processing: Fast two-dimensional median filtering 2. Segmentation: § ISODATA § Fuzzy-C-means § Active contour without edges 3. Morphological operations: Connected component labeling RESEARCH POSTER PRESENTATION DESIGN © 2011 www. Poster. Presentations. com RESULTS Sensitivity 7 -Point Checklist 90. 59% Ignoring 7 -Point Checklist 74. 78% Sensitivity Specificity 71. 80% 70. 69% Specificity Atypical Pigment Network 83. 12% 70. 16% Blue-Whitish Veil 87. 59% 76. 12% Atypical Vascular Pattern 76. 33% 71. 32% Irregular Pigmentation 86. 17% 75. 37% Irregular Streaks 84. 59% 78. 84% Irregular Dots 85. 89% 72. 27% Regression Structures 71. 37% 73. 91% ACKNOWLEDGEMENTS • • NIH grant no. IR 21 AR 057921, and grants from UH-GEAR and the Texas Learning and Computation Center at University of Houston, TX G. Argenziano et. al. Dermoscopy: a tutorial. EDRA: medical publishing and new media for skin lesion images