Colorectal Tumour Segmentation Based on 3D-UNet
- Resource Type
- Conference
- Authors
- Zhou, Wang; Mai, Ke; Peng, Hui; Xie, Qianrong; Wang, Rui
- Source
- 2021 International Conference on Computer Information Science and Artificial Intelligence (CISAI) CISAI Computer Information Science and Artificial Intelligence (CISAI), 2021 International Conference on. :430-435 Sep, 2021
- Subject
- Computing and Processing
Training
Solid modeling
Three-dimensional displays
Surgery
Predictive models
Network architecture
Planning
Automatic segmentation
3D-UNet
Attention mechanism
- Language
Colorectal cancer is one of the most common and aggressive clinical gastrointestinal cancer. The segmentation of colorectal tumor is of great significance for discovery, diagnosis, surgical planning, and treatment of colon cancer. We established 3D-UNet network on the basis of three-dimensional medical data. We added two different attention modules to the original 3D- UNet architecture at different positions to improve the accuracy of segmentation. We divided the datasets to 152 training sets and 17 testing sets at the ratio of 9:1. The experimental results of the testing sets show that our network is superior to the original 3D- UNet. The addition of attention modules results in an average segmentation accuracy increase by 1.1% and the maximum segmentation accuracy increase by 3.4%.