Description

The goal of the project is to develop a diagnostically useful machine based on image processing and recognition algorithms for atypical melanocytic lesions. Since 1994, a weekly imaging collection session has been held in the Pigmented Lesion Clinic of the Division of Dermatology, the University of British Columbia and Vancouver Hospital to digitize moles under a controlled environment. Patients were first screened by a dermatologist. Any abnormal lesions were marked and the clinical symptoms were scored. Before the lesions were excised and biopsied, An RGB colour image was obtained by a hand-held video microscopy camera, with a 20 times magnification lens. (See Fig. 1)

Figure 1. The hand-held video microscopy camera.

The RGB colour images are 512 x 486 pixels in size, with spatial resolution 0.033 mm x 0.025 mm (see Fig. 2). Each image has one or more lesions located near the centre and the lesions are surrounded by normal skin of differing hues. Other features can be observed in the images are hairs and pigments. Some images may also contain a blue marker used by the physician to designate which lesion was to be imaged. The lesion can be vary in size, shape, colour and saturation. In many cases, the margin between a lesion and the surrounding skin was clinically ill-defined.

Figure 2. A lesion image

As the first step to analyze the data set, a software program called DullRazor was implemented to remove the dark thick hairs, which can confuse the further analysis of the image. The program can be downloaded by following the link DullRazor. An automatic segmentation program to identify the lesion and the non-lesion regions has also been designed. Figure 3 shows the segmentation result for Figure 2. Currently, we are working on feature extraction algorithms. The lesion border irregularity is modelled using fractal dimensions. Other features have also been studied. Once all the features are extracted, they are used to design a classifier for normal and atypical lesions.

Figure 3. Segmentation result for Figure 1. The white line outlines the lesion border.

Principal Investigator

• Tim Lee

Research Team

• Vincent Ng
• Andrew Coldman
• David McLean
• Richard Gallagher

Funding:

• British Columbia Health Research Foundation