Pre-diagnostic digital imaging prediction model to discriminate between malignant melanoma and benign pigmented skin lesion

Abstract

Background: Malignant cutaneous melanoma is the most deadly form of skin cancer with an increasing incidence over the past decades. The final diagnosis provided is typically based on a biopsy of the skin lesion under consideration. To assist the naked-eye examination and decision on whether or not a biopsy is necessary, digital image processing techniques provide promising results.

Hypothesis and aims: The hypothesis of this study was that a computer-aided assessment tool could assist the evaluation of a pigmented skin lesion. Hence, the overall aim was to discriminate between malignant and benign pigmented skin lesions using digital image processing.

Methods: Discriminating algorithms utilizing novel well-established morphological operations and methods were constructed. The algorithms were implemented utilizing graphical programming (LabVIEW Vision). Verification was performed with reference to an image database consisting of 97 pigmented skin lesion pictures of various resolutions and light distributions. The outcome of the algorithms was analysed statistically with MATLAB and a prediction model was constructed.

Results/conclusion: The prediction model evaluates pigmented skin lesions with regards to the overall shape, border and colour distribution with a total of nine different discriminating parameters. The prediction model outputs an index score, and by using the optimal threshold value, a diagnostic accuracy of 77% in discriminating between malignant and benign skin lesions was obtained. This is an improvement compared with the naked-eye analysis performed by professionals, rendering the system a significant assistance in detecting malignant cutaneous melanoma.