Automatic multi-organ segmentation of prostate magnetic resonance images using watershed and nonsubsampled contourlet transform. (March 2016)

• Record Type: Journal Article
• Authors: Huang, Zhe; Jiang, Shan; Yang, Zhiyong; Ding, Yabin; Wang, Wei; Yu, Yan
• Is Part Of: Biomedical signal processing and control. Volume 25(2016)
• Page Start: 53
• Page End: 61
• Publication Date: 2016-03
• Subjects: Prostate magnetic resonance image -- Multi-organ segmentation -- Multi-scaled geometric image analysis -- Watershed; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2015.11.002
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: An automatic multi-organ segmentation method of prostate magnetic resonance images is proposed. The series of procedure is composed of decomposition, segmentation, and reconstitution. The decomposition and reconstitution are based on contrast à trous wavelet based wavelet transform. Marker-based watershed algorithm using texture gradient is used to implement segmentation. The main tissues in prostate magnetic resonance images can be obtained by the proposed method. Abstract: The watershed is an efficient algorithm for the segmentation of images. However, over-segmentation, which contains so many tiny regions that regions of interest cannot be identified easily, decreases the effectiveness. In this paper, pre-processing of images and the modification of watershed algorithm are both studied to restrain the over-segmentation. In the process of pre-processing, a kind of multi-scaled transform, contrast à trous wavelet based contourlet transform, is proposed and constructed to get sparse representation. In the aspect of modifying watershed, the "texture gradient" is defined, and the texture gradient is combined with marker-based watershed algorithm to reduce the number of segmented regions. The proposed method is tested by 36 prostate MR images and compared with several image segmentation algorithms; the experiment and comparison results show that the proposed method consistently restrains the number of segmented regions. The segmentation results correctly correspond to the main tissues in the images, and each tissue is integrally segmented, respectively with the elimination of small regions. The segmentation accuracy rate is 87.29%, which is higher than other methods under comparison.