Proceedings of the
The Nineteenth International Conference on Computational Intelligence and Security (CIS 2023)
December 1 – 4, 2023, Haikou, China
IIBD: Intra- and Inter-Block Densely CNN for Segmentation Application
1Department of Electronic Information and Communications Huazhong University of Science and Technology, Wuhan, China.
2School of Computer Science Guangdong University of Science and Technology Dongguan, China.
ABSTRACT
Segmentation task is an essential part of computer-aided diagnosis (CAD) systems. Despite great progress made by fully convolutional neural networks (FCNs) in recent years, existing methods fail to well segment multi-scale tumors
and accurately acquire boundaries. To solve these problems, we design a preprocessing module and present a novel end-to-end segmentation network, named IIBD (Intra- and Inter-Block Densely Connected CNN) model. Specifically, a network
comprised of intra-block and inter-block dense connections in the encoder and decoder respectively is elegantly developed to better handle the segmentation situation with multi-scale size, and a hybrid loss function combining the cross entropy loss and
Lov'asz-Softmax loss with dynamic weights is designed to learn more accurate tumor boundaries in a coarse-to-fine manner. Experimental results on the 3DIRCADb database demonstrate the superiority of the proposed model compared with classical
methods.
Keywords: Segmentation, Deep learning, Multi-scale, CT image.
Download PDF
1Department of Electronic Information and Communications Huazhong University of Science and Technology, Wuhan, China.
2School of Computer Science Guangdong University of Science and Technology Dongguan, China.
ABSTRACT
Segmentation task is an essential part of computer-aided diagnosis (CAD) systems. Despite great progress made by fully convolutional neural networks (FCNs) in recent years, existing methods fail to well segment multi-scale tumors and accurately acquire boundaries. To solve these problems, we design a preprocessing module and present a novel end-to-end segmentation network, named IIBD (Intra- and Inter-Block Densely Connected CNN) model. Specifically, a network comprised of intra-block and inter-block dense connections in the encoder and decoder respectively is elegantly developed to better handle the segmentation situation with multi-scale size, and a hybrid loss function combining the cross entropy loss and Lov'asz-Softmax loss with dynamic weights is designed to learn more accurate tumor boundaries in a coarse-to-fine manner. Experimental results on the 3DIRCADb database demonstrate the superiority of the proposed model compared with classical methods.
Keywords: Segmentation, Deep learning, Multi-scale, CT image.
Download PDF