Multi-modal Cell Segmentation based on U-Net++ and Attention Gate

Cell segmentation is one of the most fundamental tasks in the areas of medical image analysis, which assists in cell recognition and number counting. The segmentation results obtained will be poor due to the diverse cell morphology and the frequent presence of impurities in the cell pictures. In order to solve the cell segmentation which are from a competition held by Neural Information Processing Systems(NIPS), we present a network that combines attention gates with U-Net++ to segment varied sizes of cells. Using the feature filtering of the attention gate can adjust the convolution block's output, so as to improve the segmentation effect. The F1 score of our method reached 0.5874, Rank Running Time get 2.5431 seconds. 36th Conference on Neural Information Processing Systems (NeurIPS 2022).