Methods Inf Med 2009; 48(05): 399-407
DOI: 10.3414/ME9237
Original Articles
Schattauer GmbH

Automatic Brain Segmentation in Time-of-Flight MRA Images

N. D. Forkert
1   Department of Medical Informatics, University Medical Center Hamburg-Eppendorf, Hamburg, Germany
,
D. Säring
1   Department of Medical Informatics, University Medical Center Hamburg-Eppendorf, Hamburg, Germany
,
J. Fiehler
2   Department of Diagnostic and Interventional Neuroradiology, University Medical Center Hamburg-Eppendorf, Hamburg, Germany
,
T. Illies
2   Department of Diagnostic and Interventional Neuroradiology, University Medical Center Hamburg-Eppendorf, Hamburg, Germany
,
D. Möller
3   Department of Computer Engineering, Faculty of Mathematics, Informatics and Natural Sciences, University of Hamburg, Hamburg, Germany
,
H. Handels
1   Department of Medical Informatics, University Medical Center Hamburg-Eppendorf, Hamburg, Germany
› Author Affiliations
Further Information

Publication History

20 August 2009

Publication Date:
20 January 2018 (online)

Preview

Summary

Objectives: Cerebral vascular malformations might, caused by ruptures, lead to strokes. The rupture risk depends to a great extent on the individual anatomy of the vasculature. The 3D Time-of-Flight (TOF) MRA technique is one of the most commonly used non-invasive imaging techniques to obtain knowledge about the individual vascular anatomy. Unfortunately TOF images exhibit drawbacks for segmentation and direct volume visualization of the vasculature. To overcome these drawbacks an initial segmentation of the brain tissue is required.

Methods: After preprocessing of the data is applied the low-intensity tissues surrounding the brain are segmented using region growing. In a following step this segmentation is used to extract supporting points at the border of the brain for a graph-based contour extraction. Finally a consistency check is performed to identify local outliers which are corrected using non-linear registration.

Results: A quantitative validation of the method proposed was performed on 18 clinical datasets based on manual segmentations. A mean Dice coefficient of 0.989 was achieved while in average 99.56% of all vessel voxels were included by the brain segmentation. A comparison to the results yielded by three commonly used tools for brain segmentation revealed that the method described achieves better results, using TOF images as input, which are within the inter-observer variability.

Conclusion: The method suggested allows a robust and automatic segmentation of brain tissue in TOF images. It is especially helpful to improve the automatic segmentation or direct volume rendering of the cerebral vascular system.