Multi-pathways CNN for robust vascular segmentation

Multi-pathways CNN for robust vascular segmentation
Title:
Multi-pathways CNN for robust vascular segmentation
Other Titles:
Proc. SPIE 10578, Medical Imaging 2018: Biomedical Applications in Molecular, Structural, and Functional Imaging
Publication Date:
12 March 2018
Citation:
Titinunt Kitrungrotsakul, Xian-Hua Han, Xiong Wei, and Yen-Wei Chen "Multi-pathways CNN for robust vascular segmentation", Proc. SPIE 10578, Medical Imaging 2018: Biomedical Applications in Molecular, Structural, and Functional Imaging, 105781S (12 March 2018); https://doi.org/10.1117/12.2293074
Abstract:
Vascular structures are important information for education purpose, surgical planning and analysis. Blood vessels of the organ is a task that required an experienced users in order to achieve accurated result. The large variations of its structure, and properties of image are make the vessels segmentation become more and more complicate and hard to recognize even on experienced users. In this paper, we introduce a deep arti cial neural network architecture for automatically vessel segmentation of computed tomography (CT). Our network consists of multiparallel deep convolution neural networks. Each network extract the features from di erence planes to maximize a segemetation accuracy. To solve the problem of model fail to segment the clinical data which have more various constrains, we add normalization process as preprocessing. The experimental results, our network can obtain 0.879 of dice coefficient which better than stage-of-the-art methods which normally use to extract the vessels.
License type:
PublisherCopyrights
Funding Info:
Description:
ISBN:

Files uploaded:
File Size Format Action
There are no attached files.