Using Local Binary Patterns and Convolutional Neural Networks for Melanoma Detection

Skin cancer is an abnormal growth of skin cells on body parts which get more exposure to sunlight. Detection of cancer in early stages improves patient outcomes, however, manual assessment of medical cells and microscopy images is laborious work, and the results are often subjective so that the agreement between viewers can be low. In this paper, a new method is proposed to detect skin cancer signs such as asymmetry, border, colour and diameter using segmentation and region analysis. Melanoma and non-melanoma skin cancer images have been classified using region analysis, boundary, colour and size measurements. To achieve accurate and computationally efficient results, Local Binary Pattern Convolutional Neural Networks are employed. The proposed method has provided a high classification performance, achieving 0.95 accuracy rate, 0.95 sensitivity, and 0.96 specificity on the ISIC public data sets.