Automatic segmentation of 3D prostate MR images with iterative localization refinement

Accurate segmentation of the prostate gland from Magnetic Resonance (MR) images is still a challenging problem due to large variability and heterogeneity in the prostate appearance. To overcome this problem, we present a coarse-to-fine prostate segmentation approach with iterative localization refin...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Digital signal processing Jg. 98; S. 102649
Hauptverfasser: Zhou, Wenhui, Tao, Xing, Wei, Zhan, Lin, Lili
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier Inc 01.03.2020
Schlagworte:
3D U-shape network
Coarse-to-fine
Iterative localization
MR Images
Prostate segmentation
3D U-shape network
Coarse-to-fine
Prostate segmentation
Iterative localization
MR Images
ISSN:1051-2004, 1095-4333
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Accurate segmentation of the prostate gland from Magnetic Resonance (MR) images is still a challenging problem due to large variability and heterogeneity in the prostate appearance. To overcome this problem, we present a coarse-to-fine prostate segmentation approach with iterative localization refinement. Specifically, we first propose a resolution-aware 3D U-shaped network to balance the difference between the in-plane resolution and the through-plane distance. Then a case-wise loss function is introduced to alleviate the data imbalance problem and individual differences of the prostate MR images. In the inference stage, we extract a shrunk prostate region and improve the segmentation results in an iterative manner. Evaluation experiments are carried out on the MICCAI 2012 Prostate Segmentation Challenge Dataset (PROMISE12) and the NCI-ISBI 2013 Prostate Segmentation Challenge Dataset. Comparison results demonstrate that our method achieves significant improvements over the state-of-the-art approaches, and outperforms more than 290 submissions on the website of PROMISE12.
ISSN:1051-2004
1095-4333
DOI:10.1016/j.dsp.2019.102649