Skin cancer prognosis based on color matching and segmentation of pigmented skin lesion

Abstract EN

This work develops a new computerized vision of skin cancer diagnosis based on color matching of pigmented skin lesion and some parameters of ABCD method.

Initially, the clinical diagnostic criteria have been translated to mathematical concepts.

So the lesion edge detection ; symmetry ; even-symmetry ; and segmentation are computed.

Then, the suspicious images would be classified into one of three categories : benign (mole), malignant (melanoma / non-melanoma), or unknown tumor using image profile information.

The remaining malignant images (melanoma, Basal Cell Carcinoma, or Squamous Cell Carcinoma) would be further classified using matching procedure for color spectrums (Red, Yellow, Brown, Black / Gray) with lesion pigment.

The lesion image is segmented into four quarters and the matching procedure of 120 spectrums is started searching for better result with mean squared error less than 0.003.

The software has been tested over 40 classified images and it successfully re-classified 92 %.

This result could be improved if lesion quarters and / or spectrums are increased.