MDSC‐Net: A multi‐scale depthwise separable convolutional neural network for skin lesion segmentation

Accurate segmentation of the skin lesion region is crucial for diagnosing and screening diseases. However, challenging due to indistinguishable boundaries region, irregular shapes hair interference. To settle above issues, we propose a Multi-scale Depthwise Separable Convolutional Neural Network named MDSC-Net. Specifically, novel Residual Convolution Module employed in skip connection, conveying more detailed features decoder. compensate loss spatial location information down-sampling, Spatial Adaption Module. Furthermore, Decoding Fusion decoder capture contextual information. We have performed extensive experiments verify effectiveness robustness proposed network on three public benchmark datasets one polyp dataset, including ISIC-2017, ISIC-2018, PH2, Kvasir-SEG datasets. Experimental results consistently demonstrate MDSC-Net achieves superior across five popularly used evaluation criteria. The reaches high-performance segmentation, can provide important clues help doctors diagnose treat cancer early.