EOCNet:Improving Edge Omni-scale Convolution Networks for Skin Lesion Segmentation

作者：Ran Ma, Shaojie Zhang, Chao Gan, Haifeng Zhao · 年份：2020 · DOI：10.1145/3441369.3441377 · 被引用次数：1 · 研究领域：Cutaneous Melanoma Detection and Management、Nonmelanoma Skin Cancer Studies、AI in cancer detection

What makes medical imaging detection based on artificial intelligence detection, segmentation and classification to help doctors better diagnose skin cancers particularly important. Due to the different shape, size, structure, occluding hair and skin pigmentation of the lesion area, the lesion segmentation of skin cancer is challenging. We introduce a remarkably I mproving edge segmentation CNN named Edge Omni-scale Convolution Networks (EOCNet), which is represented as an encoder-decoder network. The encoder network is based on ResNet-50. The feature boundary Omni-scale module fused with the last three layers of Resnet-50 is used for the decoder. Experiments on Skin Lesion segmentation dataset achieve excellent performance.