WBC-Net: A white blood cell segmentation network based on UNet plus plus and ResNet

The counting and identification of white blood cells (WBCs, i.e., leukocytes) in blood smear images play a crucial role in the diagnosis of certain diseases, including leukemia, infections, and COVID-19 (corona virus disease 2019). WBC image segmentation lays a firm foundation for automatic WBC counting and identification. However, automated WBC image segmentation is challenging due to factors such as background complexity and variations in appearance caused by histological staining conditions. To improve WBC image segmentation accuracy, we propose a deep learning network called WBCNet, which is based on UNet++ and ResNet. Specifically, WBC-Net designs a context-aware feature encoder with residual blocks to extract multi-scale features, and introduces mixed skip pathways on dense convolutional blocks to obtain and fuse image features at different scales. Moreover, WBC-Net uses a decoder incorporating convolution and deconvolution to refine the WBC segmentation mask. Furthermore, WBC-Net defines a loss function based on cross-entropy and the Tversky index to train the network. Experiments on four image datasets show that the proposed WBC-Net achieves better WBC segmentation performance than several state-of-the-art methods. (C) 2020 Elsevier B.V. All rights reserved.

Automated summary

WBCNet, a deep learning network based on UNet++ and ResNet, improves the accuracy of white blood cell image segmentation by extracting and fusing multi-scale features using techniques like a decoder and residual blocks, achieving better segmentation performance.