An Enhanced Feature Extraction Network for Medical Image Segmentation

The major challenges for medical image segmentation tasks are complex backgrounds and fuzzy boundaries. In order to reduce their negative impacts on tasks, we propose an enhanced feature extraction network (EFEN), which is based U-Net. Our designed with the structure of re-extraction strengthen ability. process decoding, use improved skip-connection, includes positional encoding a cross-attention mechanism. By embedding information, absolute information relative between organs can be captured. Meanwhile, useful will strengthened useless weakened by using finely identify features each skip-connection cause in decoding have less noise effect object boundaries images. Experiments CVC-ClinicDB, task1 from ISIC-2018, 2018 Data Science Bowl challenge dataset demonstrate that EFEN outperforms U-Net some recent networks. For example, our method obtains 5.23% 2.46% DSC improvements compared CVC-ClinicDB respectively. Compared works, such as DoubleU-Net, obtain 0.65% 0.3%