北京理工大学 增强现实智能医疗实验室

Abstract—Image-based cell counting is a challenging task and has a wide range of clinical applications such as biomedical diagnosis and pathological analysis. In this paper, we proposed a new deep learning network structure for cell counting based on regression. First, to overcome uneven and overlap distribution of cells, we designed a dual attention U-Net (DAU-Net), which combines U-Net with spatial and channel attention to provide rich global information. Second, we designed an instance-batch normalization method to alleviate the generalization error by data augmentation, so that our model can achieve good results on data sets with different volumes. We evaluated our method on three public benchmark datasets: synthetic fluorescence microscopy dataset, human subcutaneous adipose tissue dataset, and Dublin cell counting dataset. Results showed that our method achieved satisfactory results on these three datasets.