Melanoma is one of the most deadly skin cancers and amounts for ∼79% of skin cancer deaths. Early detection and timely therapeutic action can reduce mortality owing to melanoma. In this study, we demonstrate the feasibility of our in-house skin image classification framework, trained based on a library of normal as well as pathological skin images, for automatic feature extraction and detection of melanoma. The described framework begins with active contour segmentation the skin images followed by extraction of both color and texture features from the segmented image and employs a neural network classifier to for trained identification of melanoma cases. Training and testing was conducted using a 10-fold cross validation strategy and led to 88.06% ± 1.65% accuracy in classification of melanoma images.

This content is only available via PDF.
You do not currently have access to this content.