Abstract:

Skin cancer incidence is increasing across the United Kingdom and Europe. While there are many different forms of skin cancer, the most dangerous type is known as malignant melanoma. Melanoma is known to be associated with unprotected exposure to ultraviolet radiation, with people with fair skin and freckles being at particularly high risk. Thankfully, early detection of this cancer can improve the cure rate but if allowed to progress unhindered, melanoma can metastasise and cause death. There is a relative paucity of skilled physicians worldwide with the ability to reliably and sensitively make the diagnosis of melanoma which makes computer-aided diagnostics such an enticing field within dermatology.

Conventionally, lesions that are suspected to be cancerous are biopsied (i.e. surgically excised) and put under the microscope for histological diagnosis. This is an expensive and time-consuming procedure which requires access to an individual capable of performing the procedure as well as a laboratory facilities and pathologist to examine the tissue sample. Instead, it is also possible to take image samples of the lesion and then examine the image for signs of skin cancer.

Much of the difficulty in image processing is the pre-processing required to separate healthy skin from the lesion. This is performed by manually drawing borders around the lesion on the image. While this seems a simple solution, it is time-consuming and operator-dependent.

While not a novel approach, we propose using Bayesian techniques to perform image detection and edge detection. This will obviate the need for human intervention when processing the lesion, an advantage that can make computer-aided diagnostics a far cheaper, and far more appealing alternative when clinical diagnosis by a dermatologist is unavailable. Moreover, in many settings (e.g. low-resource environments) there will be far more lesions than a single dermatologist is able to review within a reasonable amount of time. By using the above technology, it is possible to 'triage' lesions (i.e. it is possible to determine by algorithm which are the most suspicious looking lesions), thereby enabling the dermatologist to spend his or her time reviewing those lesions which are most likely to warrant treatment.

The triaging application of this technology may be used to teach trainees the characteristics of a suspicious lesion and may therefore double up as both an efficiency and teaching tool.

Keywords:

dermatology, computer-aided diagnostics, bayesian, skin cancer