CEC Theses and Dissertations

Title

An Analysis of Spectral Selectivity on Edge Detection Algorithms for a Non-invasive Identification of Skin Cancer

Date of Award

2007

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Computing Technology in Education (DCTE)

Department

Graduate School of Computer and Information Sciences

Advisor

Sumitra Mukherjee

Committee Member

Maxine S. Cohen

Committee Member

Michael J. Laszlo

Abstract

Skin cancer is the number one form of cancer in humans today. Most moles or lesions are non-cancerous (defined as benign); however, a small percentage may actually be cancerous and can ultimately be fatal. A doctor has to draw from his or her own experience and manually inspect lesions for characteristics of skin cancer during a patient's medical exam to determine whether a lesion is cancerous or benign. Being able to make this distinction could prevent serious life-threatening conditions from going undiagnosed. A review of existing literature finds that most of the current research is now focused on using computer imagery to assist the doctor in this evaluation. The basic steps include capturing the image, defining the image (e.g., shape), performing any enhancements necessary to the image (e.g., hair removal), and finally analyzing and storing the characteristics of that image (e.g., asymmetry, border, dimensions, etc.). The image has a shape that will need to be defined and this is accomplished through a process known as edge detection. Edge detection algorithms identify and locate discontinuities in the pixel intensities of an image. The discontinuities are typically associated with abrupt changes in pixel intensity values that characterize the boundaries of the objects. Since current research depends so much on computer imagery and accurate edge detection, new research should focus on what can be done to enhance the image for edge detection analysis. This dissertation focused on analyzing images of skin lesions with filtered light to determine if there were visible or non-visible characteristics of potentially cancerous lesions that the human eye could not see. This was accomplished by measuring the affect that different wavelength filters (spectral selectivity) had on lesion parameters. While this was only a small piece of the problem, it was an important building block necessary for an automated visual inspection system. Serious life threatening conditions could be better diagnosed if there was a certified visual inspection system that could distinguish benign lesions from cancerous lesions. This dissertation along with the existing body of research established a foundation on which a future system could be designed and deployed that would aid patients as well as doctors by providing additional information that can result in an improved diagnosis.

This document is currently not available here.

  Link to NovaCat

Share

COinS