About cookies on this site Our websites require some cookies to function properly (required). In addition, other cookies may be used with your consent to analyze site usage, improve the user experience and for advertising. For more information, please review your options. By visiting our website, you agree to our processing of information as described in IBM’sprivacy statement. To provide a smooth navigation, your cookie preferences will be shared across the IBM web domains listed here.
Publication
IBM J. Res. Dev
Paper
Skin lesion segmentation using deep convolution networks guided by local unsupervised learning
Abstract
Automatic localization of skin lesions within dermoscopy images is a crucial step toward developing a decision support system for skin cancer detection. However, segmentation of the lesion image can be challenging, as these images possess various artifacts distorting the uniformity of the lesion area. Recently, deep convolution learning-based techniques have drawn great attention for pixel-wise image segmentation. These deep networks produce coarse segmentation, and convolutional filters and pooling layers result in segmentation of a skin lesion at a lower resolution than the original skin image. To overcome these drawbacks, we have proposed a superpixel-based fine-tuning strategy to effectively utilize the characteristics of the skin image pixels to accurately extract the border of the lesion. Our proposed approach not only learns a global map for skin lesions, but also acquires the local contextual information, such as lesion boundary. It can, therefore, accurately segment lesions within a given skin image, even in the presence of fuzzy boundaries and complex textures. To evaluate the performance of our proposed method, experiments have been conducted using the 2016 International Symposium on Biomedical Imaging dataset, and these experiments suggest the effectiveness of the proposed method.