24 March 2016 Automated torso organ segmentation from 3D CT images using conditional random field
Author Affiliations +
Abstract
This paper presents a segmentation method for torso organs using conditional random field (CRF) from medical images. A lot of methods have been proposed to enable automated extraction of organ regions from volumetric medical images. However, it is necessary to adjust empirical parameters of them to obtain precise organ regions. In this paper, we propose an organ segmentation method using structured output learning which is based on probabilistic graphical model. The proposed method utilizes CRF on three-dimensional grids as probabilistic graphical model and binary features which represent the relationship between voxel intensities and organ labels. Also we optimize the weight parameters of the CRF using stochastic gradient descent algorithm and estimate organ labels for a given image by maximum a posteriori (MAP) estimation. The experimental result revealed that the proposed method can extract organ regions automatically using structured output learning. The error of organ label estimation was 6.6%. The DICE coefficients of right lung, left lung, heart, liver, spleen, right kidney, and left kidney are 0.94, 0.92, 0.65, 0.67, 0.36, 0.38, and 0.37, respectively.
© (2016) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Yukitaka Nimura, Yuichiro Hayashi, Takayuki Kitasaka, Kazunari Misawa, Kensaku Mori, "Automated torso organ segmentation from 3D CT images using conditional random field", Proc. SPIE 9785, Medical Imaging 2016: Computer-Aided Diagnosis, 97853M (24 March 2016); doi: 10.1117/12.2214845; https://doi.org/10.1117/12.2214845
PROCEEDINGS
6 PAGES


SHARE
Back to Top