TY - JOUR
T1 - A Feature-based learning framework for accurate prostate localization in CT images
AU - Liao, Shu
AU - Shen, Dinggang
N1 - Funding Information:
Manuscript received August 16, 2011; revised March 13, 2012; accepted March 31, 2012. Date of publication April 9, 2012; date of current version July 18, 2012. This work was supported in part by the National Institutes of Health (NIH), under Grant CA140413 and the Department of Radiation Oncology, UNC, Chapel Hill, under support of NIH, under Grant R01 RR018615 and Grant R44/43 CA119571, and in part by the National Basic Research Program of China (973 Program) under Grant 2010CB832505 and NSFC Grant 61075010. The Associate Editor coordinating the review of this manuscript and approving it for publication was Prof. Mark G. Hua.
PY - 2012
Y1 - 2012
N2 - Automatic segmentation of prostate in computed tomography (CT) images plays an important role in medical image analysis and image-guided radiation therapy. It remains as a challenging problem mainly due to three issues: 1) the image contrast between the prostate and its surrounding tissues is low in prostate CT images and no obvious boundaries can be observed; 2) the unpredictable prostate motion causes large position variations of the prostate in the treatment images scanned at different treatment days; and 3) the uncertainty of the existence of bowel gas in treatment images significantly changes the image appearance even for images taken from the same patient. To address these issues, in this paper we propose a feature-based learning framework for accurate prostate localization in CT images. The main contributions of the proposed method lie in the following aspects. 1) Anatomical features are extracted from input images and adopted as signatures for each voxel. The most robust and informative features are identified by the feature selection process to help localize the prostate. 2) Regions with salient features but irrelevant to the localization of prostate, such as regions filled with bowel gas, are automatically filtered out by the proposed method. 3) An online update mechanism is adopted to adaptively combine both population information and patient-specific information to localize the prostate. The proposed method is evaluated on a CT prostate dataset of 24 patients to localize the prostate, where each patient has more than 10 longitudinal images scanned at different treatment times. It is also compared with several state-of-the-art prostate localization algorithms in CT images, and the experimental results demonstrate that the proposed method achieves the highest localization accuracy among all the methods under comparison.
AB - Automatic segmentation of prostate in computed tomography (CT) images plays an important role in medical image analysis and image-guided radiation therapy. It remains as a challenging problem mainly due to three issues: 1) the image contrast between the prostate and its surrounding tissues is low in prostate CT images and no obvious boundaries can be observed; 2) the unpredictable prostate motion causes large position variations of the prostate in the treatment images scanned at different treatment days; and 3) the uncertainty of the existence of bowel gas in treatment images significantly changes the image appearance even for images taken from the same patient. To address these issues, in this paper we propose a feature-based learning framework for accurate prostate localization in CT images. The main contributions of the proposed method lie in the following aspects. 1) Anatomical features are extracted from input images and adopted as signatures for each voxel. The most robust and informative features are identified by the feature selection process to help localize the prostate. 2) Regions with salient features but irrelevant to the localization of prostate, such as regions filled with bowel gas, are automatically filtered out by the proposed method. 3) An online update mechanism is adopted to adaptively combine both population information and patient-specific information to localize the prostate. The proposed method is evaluated on a CT prostate dataset of 24 patients to localize the prostate, where each patient has more than 10 longitudinal images scanned at different treatment times. It is also compared with several state-of-the-art prostate localization algorithms in CT images, and the experimental results demonstrate that the proposed method achieves the highest localization accuracy among all the methods under comparison.
KW - Image-guided radiation therapy
KW - Learning-based framework
KW - Online update mechanism
KW - Prostate localization
UR - http://www.scopus.com/inward/record.url?scp=84864142669&partnerID=8YFLogxK
U2 - 10.1109/TIP.2012.2194296
DO - 10.1109/TIP.2012.2194296
M3 - Article
C2 - 22510948
AN - SCOPUS:84864142669
SN - 1057-7149
VL - 21
SP - 3546
EP - 3559
JO - IEEE Transactions on Image Processing
JF - IEEE Transactions on Image Processing
IS - 8
M1 - 6179993
ER -