Date of Award
Bachelor of Science Honours
School of Computer and Security Science
Faculty of Computing, Health and Science
Dr Alfred tan
Dr Martin Masek
The world is faced with a rapidly increasing number of skin cancers every year. Melanoma is the most deadly type of skin cancer though it can be treated if it has been detected at an early stage. However, there is a shortage of dermatologists in rural areas. The increasing number of camera phones, together with improved coverage in rural areas gives some potential for tele-dermatology, whereby people with no local access to a dermatologist can send images of suspicious skin lesions to an expert for assessment. Merely relaying images to a human expert solves only part of the problem, there is still an acute shortage of experts whose time is limited. Computer Assisted Diagnosis (CADi) of lesions promises to reduce the workload of dermatologists by acting as an assistant. Current skin lesion CADi systems employ algorithms that are designed to run on a computer at a clinic. These clinic-based systems are limited when it comes to te/e-dermatology as they rely on a suitable quality image being sent in, and need to process a large number of arriving images. An alternative to this process, afforded by the growing capabilities of mobile phones, is to do some of the CADi processing on the phone which was used to take the image. This has the potential advantage that images can be evaluated for quality on the patient's side, making it more convenient to take another image, rather than waiting for the clinic's assessment. Distributing the processing to the patient's phone also eases the workload on the clinic's machine. The first step towards implementing skin lesion CADi is the segmentation of lesions from the image background; therefore for a mobile phone to perform CADi it is a pre-requisite that it would be able to perform this step. The study seeks to determine, for an existing skin lesion segmentation algorithm, whether it is practical to adapt for mobile phone use given the limitations of the mobile camera's low resolution. The chosen algorithm depends on an edge detection step, and so an investigation will be made into edge detectors. Edge detectors are sensitive to their parameters, pixel size and lighting conditions - thus the parameters published for clinic-based systems which rely on high resolution cameras and custom lighting can not be expected ideal for mobile phone use. Experiments have shown that the approach from Xu et. al. (1999) can only apply on some types of images, which have unique background colour and distinctive from the lesion (foreground) colour.
Hua, K. L. (2007). A comparison of edge detection methods for segmentation of skin lesions in mobile-phone-quality images. https://ro.ecu.edu.au/theses_hons/1275