Jointly segmenting prostate zones in 3D MRIs by globally optimized coupled level-sets

Jing Yuan, Eranga Ukwatta, Wu Qiu, Martin Rajchl, Yue Sun, Xue-Cheng Tai, Aaron Fenster

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review

10 Citations (Scopus)

Abstract

It is of great interest in image-guided prostate interventions and diagnosis of prostate cancer to accurately and efficiently delineate the boundaries of prostate, especially its two clinically meaningful sub-regions/zones of the central gland CZ and the peripheral zone PZ, in the given magnetic resonance MR images. We propose a novel coupled level-sets/contours evolution approach to simultaneously locating the prostate region and its two sub-regions, which properly introduces the recently developed convex relaxation technique to jointly evolve two coupled level-sets in a global optimization manner. Especially, in contrast to the classical level-set methods, we demonstrate that the two coupled level-sets can be simultaneously moved to their globally optimal positions at each discrete time-frame while preserving the spatial inter-surface consistency; we study the resulting complicated combinatorial optimization problem at each discrete time evolution by means of convex relaxation and show its global and exact optimality, for which we introduce the novel coupled continuous max-flow model and demonstrate its duality to the investigated convex relaxed optimization problem with the region constraint. The proposed coupled continuous max-flow model naturally leads to a new and efficient algorithm, which enjoys great advantages in numerics and can be readily implemented on GPUs. Experiments over 10 T2-weighted 3D prostate MRIs, by inter- and intra-operator variability, demonstrate the promising performance of the proposed approach.
Original languageEnglish
Title of host publicationEMMCVPR 2013: Proceedings of the 9th International Conference on Energy Minimization Methods in Computer Vision and Pattern Recognition
EditorsAnders Heyden
Place of PublicationBerlin
PublisherSpringer
Pages12-25
Number of pages14
ISBN (Print)978-3-642-40394-1
DOIs
Publication statusPublished - 2013
Externally publishedYes

Publication series

NameLecture Notes in Computer Science (LNCS)
Volume8081

Fingerprint

Dive into the research topics of 'Jointly segmenting prostate zones in 3D MRIs by globally optimized coupled level-sets'. Together they form a unique fingerprint.

Cite this