Cervical cell nuclei segmentation based on GC-UNet
Enguang Zhang,
Rixin Xie,
Yuxin Bian,
Jiayan Wang,
Pengyi Tao,
Heng Zhang,
Shenlu Jiang
Affiliations
Enguang Zhang
School of Computer Science and Engineering, Macau University of Science and Technology, Macau, China; Zhuhai College of Science and Technology, Zhuhai, China
Rixin Xie
School of Computer Science and Engineering, Macau University of Science and Technology, Macau, China
Yuxin Bian
School of Computer Science and Engineering, Macau University of Science and Technology, Macau, China
Jiayan Wang
School of Computer Science and Engineering, Macau University of Science and Technology, Macau, China
Pengyi Tao
School of Computer Science and Engineering, Macau University of Science and Technology, Macau, China
Heng Zhang
Faculty of Education, The University of Hong Kong, Pokfulam Road, Hong Kong, China
Shenlu Jiang
School of Computer Science and Engineering, Macau University of Science and Technology, Macau, China; Corresponding author.
Cervical cancer diagnosis hinges significantly on precise nuclei segmentation at early stages, which however, remains largely elusive due to challenges such as overlapping cells and blurred nuclei boundaries. This paper presents a novel deep neural network (DNN), the Global Context UNet (GC-UNet), designed to adeptly handle intricate environments and deliver accurate cell segmentation.At the core of GC-UNet is DenseNet, which serves as the backbone, encoding cell images and capitalizing on pre-existing knowledge. A unique context-aware pooling module, equipped with a gating model, is integrated for effective encoding of ImageNet pre-trained features, ensuring essential features at different levels are retained. Further, a decoder grounded in a global context attention block is employed to foster global feature interaction and refine the predicted masks.
