Home
Research Team
Projects
Publications
Contact Us
 
Malignant melanoma poses a significant risk to Canadians and continues to rise. Therapy for advanced melanoma remains poor, with a five-year survival rate of 50% or less. However if melanoma is detected early, the five-year survival rate is around 95%. Therefore, early diagnosis of melanoma is critical so that it can be completely excised surgically while it is still localized. Our goal is to develop an automatic screening system for evaluating suspicious pigmented skin lesions including both melanocytic (moles) and non-melanocytic lesions, which can be used to help family practitioners (FPs) and other health care professionals determine whether or not a patient should undergo an invasive skin biopsy or be referred to a specialist dermatologist. Additionally we hope that the system will be used by experts as a “second reader” to help expert dermatologists to diagnose malignant melanomas and atypical moles.

Clinical diagnostic methods involve visual inspection and simple diagnosis based on common features: Asymmetry, Border irregularity, Colour and Diameter (ABCD). Experts may also examine a skin lesion under high magnification obtained using a dermoscope (a hand-held magnifying device using polarized light), and consider further features for classification, especially those involving texture and textural patterns. Our goal is to automatically quantify the textural information in skin lesions and incorporate these texture features into an automatic diagnosis system.

We at the SFU's School of Computing Science, Medical Image Computing Analysis Laboratory, are working closely with the Skin Care and Dermatology Centre at UBC to achieve these goals, through 3 major efforts:

  1. Developing and evaluating a novel low-cost dermoscope system which will provide the necessary high-resolution images for our automated diagnosis system.
  2. Developing machine learning and image processing techniques to identify major features used for classification of images
  3. Developing methods to automatically, and reliably, process a large number of skin lesion images.

 

Home Research Team Projects Publications Contacts
Last updated: May 4, 2010