An Improved Multi-Scale Feature Fusion for Skin Lesion Segmentation

Accurate segmentation of skin lesions is still a challenging task for automatic diagnostic systems because the significant shape variations and blurred boundaries lesions. This paper proposes multi-scale convolutional neural network, REDAUNet, based on UNet3+ to enhance network performance practical applications in segmentation. First, employs new encoder module composed four feature extraction layers through two cross-residual (CR) units. configuration allows extract deep semantic information while avoiding gradient vanishing problems. Subsequently, lightweight efficient channel attention (ECA) introduced during encoder’s stage. The assigns suitable weights channels learning effectively captures inter-channel interaction information. Finally, densely connected atrous spatial pyramid pooling (DenseASPP) inserted between decoder paths. integrates dense connections ASPP, as well fusion, recognize varying sizes. experimental studies this were constructed public lesion datasets, namely, ISIC-2018 ISIC-2017. results show that our model more accurate segmenting different shapes achieves state-of-the-art In comparison UNet3+, proposed REDAUNet shows improvements 2.01%, 4.33%, 2.68% Dice, Spec, mIoU metrics, respectively. These suggest well-suited can be employed computer-aided systems.