A Convolutional Neural Network for Skin Lesion Segmentation Using Double U-Net Architecture

Skin lesion segmentation plays a critical role in the precise and early detection of skin cancer via recent frameworks. The prerequisite for any computer-aided diagnosis system is accurate malignancy. To achieve this, specialized image analysis technique must be used separation cancerous parts from important healthy skin. This procedure called Dermatography. Researchers have often multiple techniques images, but, because their low accuracy, most these methods turned out to at best, inconsistent. Proper clinical treatment involves sensitivity surgical process. A high accuracy rate therefore paramount importance. generalized robust model needed accurately assess segment lesions. In this regard, novel approach named Double U-Net has been proposed provide necessary strength Robustness. process uses two architectures stacked upon each other with ASPP which squeeze resolution redundant information. paper, we trained architecture on PH2 dataset was evaluated test, ISIC-2016 HAM datasets. Evaluation information shows achieved DSC 0.9551 test dataset, 0.8104 0.7645 dataset. Analyses show results comparable recently available state-of-the-art techniques.