Masters Theses

Hanzheng Wang

Keywords and Phrases

Malignant Melanoma; Watershed Algorithm

Abstract

"Automatic lesion segmentation is an important part of computer-based skin cancer detection. A watershed algorithm was introduced and tested on benign and melanoma images. The average of three dermatologists' manually drawn borders was compared as the benchmark. Hair removing, black border removing and vignette removing methods were introduced in preprocessing steps. A new lesion ratio estimate was added to the merging method, which was determined by the outer bounding box ratio. In postprocessing, small blob removing and border smoothing using a peninsula removing method as well as a second order B-Spline smoothing method were included. A novel threshold was developed for removing large light areas near the lesion boundary. A supervised neural network was applied to cluster results and improve the accuracy, classifying images into three clusters: proper estimate, over-estimate and under-estimate. Comparing to the manually drawn average border, an overall of 11.12% error was achieved. Future work will involve reducing peninsula-shaped noise and looking for other reliable features for the classifier"--Abstract, page iii.

Advisor(s)

Moss, Randy Hays, 1953-

Committee Member(s)

Stanley, R. Joe
Stoecker, William V.

Department(s)

Electrical and Computer Engineering

Degree Name

M.S. in Electrical Engineering

Publisher

Missouri University of Science and Technology

Publication Date

2009

Pagination

viii, 58 pages

Note about bibliography

Includes bibliographical references (pages 55-57).

Rights

© 2009 Hanzheng Wang, All rights reserved.

Document Type

Thesis - Open Access

File Type

text

Language

English

Subject Headings

Melanoma -- Diagnosis
Skin -- Cancer -- Diagnosis
Image segmentation
Image processing -- Computer programs
Neural networks (Computer science)

Thesis Number

T 10567

Print OCLC #

908250220

Electronic OCLC #

908262055

Download

Share

 
COinS