Share Email Print
cover

Proceedings Paper

Prostate segmentation in 3D TRUS using convex optimization with shape constraint
Author(s): Wu Qiu; Jing Yuan; Eranga Ukwatta; David Tessier; Aaron Fenster
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

An efficient and accurate segmentation of 3D end-firing transrectal ultrasound (TRUS) images plays a central role in the planning and treatment of 3D TRUS guided prostate biopsy. In this paper, we propose a novel convex optimization based approach to delineate prostate boundaries from 3D TRUS images. The technique makes use of the approximate rotational symmetry of prostate shapes and reduces the original 3D segmentation problem to a sequence of simple 2D segmentation sub-problems by means of rotationally reslicing the 3D TRUS images. In practice, this significantly decreases the computational load, facilitates introducing learned shape information and improves segmentation efficiency and accuracy. For each 2D resliced frame, we introduce a new convex optimization based contour evolution method to locate the 2D slicewise prostate boundary subject to the additional shape constraint. The proposed contour evolution method provides a fully time implicit scheme to move the contour to its globally optimal position at each discrete time, which allows a large evolving time step-size to accelerate convergence. Moreover, the proposed algorithm is implemented on a GPU to achieve a high performance. Quantitative validations on twenty 3D TRUS patient prostate images demonstrate that the proposed approach can obtain a DSC of 93:7 ± 2:5%, a sensitivity of 91:2 ± 3:1%, a MAD of 1:37 ± 0:3mm, and a MAXD of 3:02 ± 0:44mm. The mean segmentation time for the dataset was 18:3 ± 2:5s, in addition to 25s for initialization. Our proposed method exhibits the advantages of accuracy, efficiency and robustness compared to the level set and active contour based methods.

Paper Details

Date Published: 13 March 2013
PDF: 7 pages
Proc. SPIE 8669, Medical Imaging 2013: Image Processing, 866943 (13 March 2013); doi: 10.1117/12.2006836
Show Author Affiliations
Wu Qiu, Robarts Research Institute (Canada)
Jing Yuan, Robarts Research Institute (Canada)
Eranga Ukwatta, Robarts Research Institute (Canada)
The Univ. of Western Ontario (Canada)
David Tessier, Robarts Research Institute (Canada)
Aaron Fenster, Robarts Research Institute (Canada)
The Univ. of Western Ontario (Canada)


Published in SPIE Proceedings Vol. 8669:
Medical Imaging 2013: Image Processing
Sebastien Ourselin; David R. Haynor, Editor(s)

© SPIE. Terms of Use
Back to Top