A Non-invasive Automatic Skin Cancer Detection System for Characterizing Malignant Melanoma from Seborrheic Keratosis

Abstract

Due to the complexity of skin cancer treatment at later stages, the investigation of an efficient non-invasive automated system can help in guiding diagnosis. This paper proposes a non-invasive automatic system for characterizing malignant melanoma from seborrheic keratosis (BKL) using pixel-based segmentation and feature extraction techniques. The proposed system utilizes the pixel-based features to capture the main characteristics that discriminate BKL and malignant melanoma (MEL). The pixel-based technique enabled single-pixel distributions for color and texture that results in good discrimination of pigmented skin lesions from unaffected skin regions in the processed image. In the experimental results, the obtained characterization result using gradient boosted trees (GBT) is promising and outperformed other state-of-the-art techniques, which had an accuracy equaled to 97.5%, Dice measure equaled to 98.5%, sensitivity equaled to 98.3%, and specificity equaled to 92.1%.