MLRNet: Skin Lesion Segmentation using Hybrid Gaussian Guided Filter with CNN

Abstract

Recently, the worldwide population has been suffering with skin cancer issues due to the unpredictable radiation levels of environment, which may lead to cause of death. Therefore, the early detection of skin cancer through accurate skin lesion segmentation can save the human life all around the world. Deep Learning has demonstrated improved performance in several modalities of biological image analysis in recent years. This work adopted the multi-layer residual convolutional neural network (MLRNet) for skin cancer segmentation. Initially, the test skin lesion image is preprocessed by using hybrid gaussian guided image filter (HGGIF) in discrete wavelet transform (DWT) domain, which removes the noise and various artifacts from the skin lesion images. Further, the filtered image is applied to MLRNet, which generates the accurate segmentation map. The simulations are carried out on both ISIC-2019 and PH <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> datasets. The proposed method outperformed in both subjective and objective performances as compared to conventional deep learning approaches. Keywords- Skin Lesion Segmentation, Deep learning, Multi-Layer Residual Convolutional Neural Network, Gaussian filter, Guided Image Filter