TY - GEN
T1 - Boundary-Aware Network for Kidney Tumor Segmentation
AU - Hu, Shishuai
AU - Zhang, Jianpeng
AU - Xia, Yong
N1 - Publisher Copyright:
© 2020, Springer Nature Switzerland AG.
PY - 2020
Y1 - 2020
N2 - Segmentation of the kidney and kidney tumors using computed tomography (CT) is a crucial step in related surgical procedures. Although many deep learning models have been constructed to solve this problem, most of them ignore the boundary information. In this paper, we propose a boundary-aware network (BA-Net) for kidney and kidney tumor segmentation. This model consists of a shared 3D encoder, a 3D boundary decoder, and a 3D segmentation decoder. In contrast to existing boundary-involved methods, we first introduce the skip connections from the boundary decoder to the segmentation decoder, incorporating the boundary prior as the attention that indicates the error-prone regions into the segmentation process, and then define the consistency loss to push both decoders towards producing the same result. Besides, we also use the strategies of multi-scale input and deep supervision to extract hierarchical structural information, which can alleviate the issues caused by variable tumor sizes. We evaluated the proposed BA-Net on the kidney tumor segmentation challenge (KiTS19) dataset. The results suggest that the boundary decoder and consistency loss used in our model are effective and the BA-Net is able to produce relatively accurate segmentation of the kidney and kidney tumors.
AB - Segmentation of the kidney and kidney tumors using computed tomography (CT) is a crucial step in related surgical procedures. Although many deep learning models have been constructed to solve this problem, most of them ignore the boundary information. In this paper, we propose a boundary-aware network (BA-Net) for kidney and kidney tumor segmentation. This model consists of a shared 3D encoder, a 3D boundary decoder, and a 3D segmentation decoder. In contrast to existing boundary-involved methods, we first introduce the skip connections from the boundary decoder to the segmentation decoder, incorporating the boundary prior as the attention that indicates the error-prone regions into the segmentation process, and then define the consistency loss to push both decoders towards producing the same result. Besides, we also use the strategies of multi-scale input and deep supervision to extract hierarchical structural information, which can alleviate the issues caused by variable tumor sizes. We evaluated the proposed BA-Net on the kidney tumor segmentation challenge (KiTS19) dataset. The results suggest that the boundary decoder and consistency loss used in our model are effective and the BA-Net is able to produce relatively accurate segmentation of the kidney and kidney tumors.
KW - Boundary detection
KW - Computed tomography
KW - Deep learning
KW - Kidney and kidney tumor segmentation
UR - http://www.scopus.com/inward/record.url?scp=85092693183&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-59861-7_20
DO - 10.1007/978-3-030-59861-7_20
M3 - 会议稿件
AN - SCOPUS:85092693183
SN - 9783030598600
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 189
EP - 198
BT - Machine Learning in Medical Imaging - 11th International Workshop, MLMI 2020, Held in Conjunction with MICCAI 2020, Proceedings
A2 - Liu, Mingxia
A2 - Lian, Chunfeng
A2 - Yan, Pingkun
A2 - Cao, Xiaohuan
PB - Springer Science and Business Media Deutschland GmbH
T2 - 11th International Workshop on Machine Learning in Medical Imaging, MLMI 2020, held in conjunction with the 23rd International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2020
Y2 - 4 October 2020 through 4 October 2020
ER -