Skin Lesion Segmentation based on Integrating EfficientNet and Residual block into U-Net Neural Network

Skin lesion segmentation is an important step in computer aided diagnosis for automated melanoma diagnosis. However, in the field of medical images analysis, skin lesion segmentation from dermoscopic images is stilla challenging task be-cause of presence of various artifacts, blurring and irregular edges of the lesion. This paper proposes an efficient deep learning-based approach for skin lesion segmentation. Particularly, the paper proposes an improved version of the U-Net to perform skin lesion segmentation tasks. To this end, we propose to utilize Effi-cientNetB4 in encoder part of the original U-Net. In addition, the decoder part of the proposed network is constructed by residual block from Resnet architecture. By this way, the proposed approach could take advantages of the EfficientNet and Resnet architectures such as preserving efficient reception field size for the model, and avoiding the overfitting problem. The proposed approach is applied to segment images from ISIC 2017 and 2018 datasets. Experimental results show the desired performances of the proposed approach in terms of metrics of Dice coefficient and Jaccard indexes.