Analysis and Quantification of the Textural Information in Pigmented Skin Lesions

There are certain features of skin lesions or moles that are indicative of cancer. Texture (spatial variation of colour/intensity) or patterns is known to be one of these features. This research seeks to quantify this textural information in order to improve the performance of methods to automatically diagnose melanoma. Currently, dermatologists are very proficient at interpreting this textural information, and easily outperform automated techniques. This research will analyze how experts perceive this information, as well as the relative importance of various textural patterns and their locations. A mathematical model of texture will then be developed, attempting to incorporate the prior knowledge gained from analyzing the experts. Additionally, digitally removing occluding hair from images of skin lesions is a critical element in any automated system of melanoma diagnosis. This research will also seek to improve on existing methods of hair detection in images, as well as estimating the underlying skin color of ‘hair pixels’.

Faculty Supervisor:

Dr. M. Stella Atkins

Student:

Paul Wighton

Partner:

BC Cancer Agency

Discipline:

Computer science

Sector:

Legal

University:

Simon Fraser University