Skin Lesion Segmentation Based on Deep Learning

The precise segmentation of the skin pathology area plays an important role in the determination of the skin disease type. This paper proposed the method in Deep learning, Mask R-CNN, to segment skin diseases, and introduced K-means clustering algorithm in the pre-processing of the data set. This method can ensure the accurate labeling of the data set itself, while accurately segmenting the skin pathology area. Experiment results based on ISIC (International Skin Imaging Collaboration) data set with 23906 images demonstrate that the segmentation effect on skin image and the segmentation accuracy of its test results is as high as 91.87%. Further more, It can segment lesions within a given skin image, even in the presence of fuzzy boundaries and complex textures.